I hear all of you on this one, especially the OP. I can also tell a difference... ...Here's the deal: I grew up nearly deaf (and inching towards deafness). Sound didn't matter until a surgery gave me back full function in one ear. The other is bad, but with special equipment, can work up to 89% efficient, which, I'm told is average human quality for my age. I can only suggest that my experience of sound in such a manner has allowed me to articulate greater differences in how sound is experienced. To that I add the following:
I play on both near field speakers, as well as on a full surround system that was professionally installed and engineered with microphone field balance (connecting to mics and testing at different fields of the room). I can tell a difference. However, it's not an articulated auditory response (not necessarily pure "Hearing" of the sound). It's more in the "Feel" of the sound. I use a USB enabled receiver, with 48, 96 and 192khz sample rates in WAV. With 96k there's a touch more entry feeling than 48, and the anticipated sound feels fuller, more complete; also, the roll into the next notes (I work performances and do live audio recording for video post production archival) is more fluid. However, this is only slightly more noticeable in 192khz, or, I believe, my specs are maxed (on the receiver) at 96 and it interpolates the data down.
The old 20khz human response is based on sound room in-ear otiology, which limits the sound to what you can hear in your ears alone, but hearing is done throughout the body, and processed by multiple systems in the brain.
If you're unhappy with the mixdown, check your dithering and antialiasing modes in your DAW. Do some reading on their balances. You might find a preset that works well for you. They sort of "Fake" the feeling of the sound, much like the blur\sharpening effects of photoshop do on photos. Play around with each audio style you work with. IF you have a compression GUI that will accept pure WAV and give you effects to choose from (like apple's compressor), you can encode several files and check the balance. I would invest in a USB surround receiver and\or usb playback device, then put your original WAV or full format files on USB, along with your mixed set. Know your receiver's limits (check it out online or send a support request with your questions to the manufacturer). Play Original sound, then a mixed sound in headphones, and repeat with each mix, making notes about how much closer they get. DO the same with near field speakers. Do the same with your wide room surround. Once you get "Close Enough" for your target output, check that mix, and create a preset if possible. This will give you the best results to your preference. Audition gives you only one output at a time. AME doesn't really allow you to add effects, and doesn't allow you to tweak the downsample. You're best off with a program similar to AME, or More like Compressor, that allows you to add basic audio effects to properly mix down or compress the range, with little perceptual loss.
Most audio systems are targeted to the ear, because it is the primary sensory organ for audio response. But most also include a SubWoofer that processes frequencies and builds a "Feel" of percussion or Force into the mix based on presets in the receiver. The better your subwoofer, the more natural and forceful the feel; but don't discount the receiver's algorithm, as it sends a processed range to the subwoofer. If you play a 96khz naturally recorded sound and a 48khz naturally recorded sound (both being the same sound), you'll not notice the difference on any really explicable level. IF you are so tuned, however, you might (6-28% average of professionals) be able to vocalize that the sound just "feels" different, and you'd be right. In recording digitally, you don't record all frequencies, normally. With this in mind as a hypothesis, I ran tests.
My conclusions: With 48khz samples, some of the lower band is dropped to favor the higher bands, stretching to include them and provide headroom for editing. With 96khz, more of the lower is processed but it's still stretched up. And with 192, the lower band is almost completely covered, and there is no shift, so that it doesn't stretch to higher bands. Beyond this, I haven't seen any of them attempt to stretch. The bottom is usually about 100hz-300hz for 48k and below. With 96 it's more like 50hz-200hz.
I tested this by producing a varying tone and checking the waveform that was recorded. Everything that didn't reach a certain decibel level (about -15db) was ignored at 300hz with 48khz, and lower than that, it was down to -9db. This is where the drop comes in. I was able to pick up sound at -40db at 350hz. Upon trying to adjust where there was apparently drop, I tried to see if there was anything there to view at all, and there was nothing but signal noise. I checked my hardware by inputing the sound at full output volume into the capture device, which showed input even at 0 attenuation, then used it to input by usb into the computer across 48 and 96khz files. Same behavior. I had to attenuate to catch the sound and have the attenuation almost a quarter up to pick up anything (though not even audible). I then played the files back on my surround system, and noticed there was some vibration of the drivers on the subwoofer at a basic level (about 15 on it's volume). This is a "normal" level for basic ear hearing (according to the documentation it's the volume where you should be able to hear the sound in that size room if you have normal hearing), and didn't really "Feel" like much. I raised the volume slowly and got signal noise, but as I raised the volume I began to hear a smoother tone (up around the max, which was 40-50 on volume attenuation).
This last test proves that those frequencies are played, even if we don't exactly "Hear" them. With a 96k playback I could actually feel a bit of a smooth effect in the sound, under the sound really. Does it prove everything? No. It does hint at the engineering of the hardware and software. Maybe my DAW is trying to allow for the sharp and shrill being balanced (vocals and high instruments that have a more "cutting" feel), and it's only filling the base or lows that can be played by modern hardware.
By the way, for future reference, the sample rate with digital audio has more to do with cutting up the sound rather than the frequency; however, since sound is a wave phenomena that can be graphically represented by plot points with numeric value, cutting it up does offer a way to capture the frequencies at a given instant. The difference is that everything in between is Guessed, as there is not solid algorithm for how the sound molds in between the two plotted points. The rule of thumb is: More plotted points, more accurate graph of the sound. With a continuous tracking (not sampled but continuously captured), the sound would be a nearly perfect representation, if such a tracking method could be done. However, tracking and plotting the points would be an infinitesimal task, occurring an infinite number of times per second. Since this isn't possible, we attempt to reproduce the sound using the wave phenomena as a guide, sampling the wave at points along the propagation, with similar frequency to the wave itself. This has the following effect: we can reproduce a 48khz width of sound with some accuracy, but as the sounds are piled together across frequencies, the samples don't perfectly match up and some of the information is lost, making the transitions sharper and less curved\smooth (graphically). When you increase your kHz, you can reproduce the sound more accurately, and capture the higher frequencies, but only if, at some point, you process and adjust the capture to pick up the right point of that frequency (the high and\or low of the wave) rather than an in between. Thus, you'll almost never be perfectly accurate. You'll lose quality when you drop your kHz, mainly because you're dropping the in-between points off the graph, and you'll find that your sound really loses more "Feel". However, by dithering and adjusting the Aliasing of the drop, you can provide a curvature to the interpolation, and "Fake" the feel a little. The samples on the recorder have nothing to do with the kHz of microphone or the sounds actual frequency, other than that you can sample at the same speed as the waves, and mark the amplitudes you measure. However, nothing says those amplitudes are right on the mark for the peak and trough of the actual wave. There is already a drop in quality. Add the fact that those measurements are plotted, and then connected crudely, and you can understand that its not the frequency you are measuring, its the amplitude of each interpolated sound frequency at that particular sampling moment (and they are all split into their respective ranges with varying accuracy of numerics; your bit width). For example, I can capture a 96khz sound at 48khz, and if it is a continuous tone over two or more seconds, it will sound almost the same when interpolated. If it varies at all, it will lose that variation quality. When playing back analog audio, the amplitudes are represented for the different frequencies over continuous waves. With digital, there is a conversion to a continuous format for the drivers to represent, as the frequencies are all split up and have to be mixed, and there is some shifting or inaccuracy of the timing. This is why most say that a Vinyl LP still sounds better than a CD.