At LinuxWorld today, a couple of developers from IBM stopped by and mentioned a patch that they recently sent to Linus (which he accepted with dismay at such a long standing bug, and apparently committed to the 2.4.8 tree).
Sadly, I have forgotten the woman's name, but she indicated that there was a race condition that could sometimes cause the fs register to become corrupt, and that this problem had existing in all versions of Linux, as far as they could tell.
She indicated that the patch had a fairly innocuous description, but it might be worth understanding this problem better (although from her description, it didn't sound like a good workaround was available).
Cheers,
Jeremy
p.s. Wine was a finalist for best open source project, but lost out to KDE. Next time, the crystal plaque is ours. I can feel it.