It sounds to me that you're running into a pulldown issue, which WILL cause sync to gradually drift over time. Check your PT session setup and make sure there's no pulldown/up enabled (and subsequently, if there's a way to check this in FCP, it's, worth doing so as well). Framerates and pulldown/up are different animals - you can have two things using the identical frame rate (PAL 25, which may look just fine at first), but if one has a pulldown and the other doesn't (or they both have different pulldowns), it will throw sync regardless.
Also, more clarification would be helpful. Are you running ProTools while running FCP? Is there a reason you need to be if that's the case? Is your computer audio out feeding in the PT hardware so you can audition it on the mains, and that's why you need PT open or use the HW device via CoreAudio?
The only issue I see so far is that when you load in a printed file into FCP and it's causing a sync drift against pix, you have a pulldown/up issue or incorrect framerate at which the original print was made at (like 23.976 or 29.97 which run at 0.1% slower than 24 and 30), and that's causing the discrepancy you see in FCP with the audio print.
As to the secondary issue of the HW device and sample rate, some more clarification as to the circumstances would be helpful. It doesn't seem to figure into the equation of the problem unless I'm missing some key information (as questioned above).
If what your saying is that the HW is serving as your mains even with PT is not running (e.g. using it via CoreAudio), it's possible that by having different sample rate settings it can cause an issue where everything you hear from that device will be out of sync, but normally devices will override with the sample rate they need, unless the HW is still slaved to another device (e.g. trying to run PT while feeding CoreAudio to it at the same time). If that's the case, try setting your computer audio to 48k (which I gather is the regular session sample rates your working at). If this is the case then it's an entirely different issue.
Ultimately, further articulation about the problem would help, it's stil muddy to me what's going on and what the focus of the problem actually is.