From: "Frank Haber" 

(I don't know whether my chops are adequate to be the Voice of Hardware
here, but there are some factors.....)

o Antti hit it on the head on the overlapped I/O issue.

o Most USB drives are FAT, and most people should keep them that way,
because just ONCE they'll forget and pull it out when they shouldn't. 
Formatting big thumb drives NTFS is LOTS faster, if you like to speed-skate
on thin ice. With FAT, you're totally safe (IMHO) if you just wait for the
activity light to go out on the thumb.  No need to dismount. And most
recent Mac and xxNIX opsyses read/write FAT. (I'm waiting for the first
dufus to accidentally kick out his cached Vista kernel from a loose and
dirty USB slot.  That'll be geek humor.)

o Firewire is some kind of weird exception on 2k and XP.  And the
controller manufacturers are apparently all asleep, except for Oxford. 
Expect delayed-write errors if you regularly write 50-500MB files to
external firewire HDs.  It's really a shameful situation.  (Hint: this
would be a really good place for some programmer to insert a very
old-fashioned small-block, read-write-verify alternate copy routine in a
backup or filecopy program.  It'd be unique.  Sticky-by-drive_letter should
be good enough for this option, unless some API let's you know when your
target is IEEE 1394.)

o Overlapped/interleaved I/O is peachy keen and a good thing on recent
hardware.  If you ever have occasion to play with early UDMA33 drives, or
worse yet, the ISA bus and polled drives, watch out.  If the copy matters
(including sec_copy/Ghosting), ALWAYS place source and target ALONE on
SEPARATE IDE channels. If the target  drive is an old one, consider forcing
the DMA mode lower by using a 40-pin cable there.

--- BBBS/NT v4.01 Flag-5