Interesting-- got some conflicting responses. The question seems to depend on whether or not the driver chips are different between the two or not-- one poster seems to think they may not be-- but the majority seems to think otherwise.
Turns out both sides are right. Phillips, and Mentor graphics (AMI) make a "Dual Role" part which can act as either the host or the device. However, a lot of parts appear to be "host controllers" with no mention of any ability to operate on the device end. There are also "transciever" parts which may leave it up to the host CPU, but I kinda doubt that would be what my device is using. So if 1) my device had one of these dual-role type chips AND there are Linux drivers for it or I feel like writing one-- then perhaps there's a chance, but it's starting to look like the odds are not in favor of it.
Raises another question though-- there do seem to be "bridge" or host-to-host configurations. While it would seem to be *theoretically* possible to construct a intermediary device that will fake the PC end into thinking it is talking to a standard USB drive, the question would then be how complex would the intermediary have to be given the other end could have whatever kind of alternative driver necessary created for it? I'm not talking about using just a special cable and custom software on both ends, but a box in the middle that acts as device for both ends and passes the data through in such a way as at least the PC end of the chain doesn't need special drivers and just sees a standard USB drive...