I've just this very moment finished a very short essay for my Masters of Audio and Acoustics. It was a seminar class so we didn't really go very indepth into Ambisonics, but we did build a 16 speaker array and used max/msp to localise sounds within a sphere. 4 speakers above, four below and 8 horizontal to listener.
The class was more about the design aspects of a room and getting the acoustic response of the room and loudspeakers so we could adjust the room properly.
I was tasked with creating the sounds to be replayed through the system.
I made a bunch of sound scapes and musical pieces and placed these within co-ordinates of the sphere in max replicating the sphere of the loudspeaker array.
The main things I noticed (this was only with one or two days fiddling around in the room) was
* that panning showed up the gaps between speakers( slower pans = better movement),
* co-axial loudspeakers would be better to localise the sound,
* the directivity of the speakers at different frequencies causes localisation issues.
Some plugins are available, I have not tested these out as I've just downloaded them myself
Here's an article on mixing http://www.ambisonic.net/ambimix.html THis is more about mixing down to a physical encoder/decoder setup, I'm sure it would be much simpler in a DAW.
For mixing it was difficult as we were limited to the max patch that the lecturer was working with. It didn't allow for any automation, so all sounds were statically placed in the sphere, this made it simpler because the computer wasn't in the localisation spot and i could jump in and out to adjust. If i was mixing something that required panning or mixing to film then you would need to be in the centre of the array to achieve the best result.
One cheat we needed was that there was very little low frequencies coming from the small loudspeakers we were using. So we added a filter below 250Hz to send the sound to all loudspeakers, a similar cheat would be to use a separate subwoofer.
I think ambisonics would be a great archival medium for surround work as it stores angular information for location and not just the speaker you wish the sound to come from as in stereo/5.1/7.1/22.2, this way future speaker arrays or listening situations could use an ambisonically recorded work without the production needing remixing.