> I think there's sufficient evidence to suggest that this is the case (e.g.
> the problem does not occur if you halt the machine but do not remove power;
> the kernel is no longer running, but the data eventually makes it to disk).

I'd like to point out that this does happen also with non-SCSI systems
(think I filed a pr about this quite some time ago). I have an IDE system
which 1 out of 5 times I do "shutdown -r now" doesn't come up again because
the primary master block copy is different from the second one. Running fsck
manually fixes this immediately and no errors or warnings are shown on by
fsck.

To bad this is a remote system with no easy way to reach the console (since
it's that places' internet router as well).


Martin