It occurred to me that Richard Rauch wrote in gmane.os.netbsd.help:
> Thanks for the little script.
>
> I'm not even 100% sure that I needed it.  I (re)discovered that the USB
> drive's MS-DOS filesystem is, for some reason, put on disklabel partition
> e.  (Why not a, as with CD filesystems?)  I only tried the obvious sd0a
> prior to running your script.  Now I have sd0e mounted.

There is some logic behind that:

For all purposes, an USB key is just the same as any other (SCSI) disk,
so it gets exactly the same treatment as any other SCSI disk. If no disklabel
can be found, it will generate a disklabel for you - and the first available
partition is 'e'. 'd' is the whole, raw disk, and it's already in use
anyway, 'c' would be the NetBSD portion of that, but since there is none,
it's empty, 'b' would be very inappropriate, because 'b' is reserved for
swap, and 'a' is reserved for a NetBSD root partition, which isn't the case
either.

so 'e' is actually quite logical. The logic *ends* with the fact that these
subpartition letters are often abused for very different purposes on non-disk
devices. Like on floppies, where they are used to indicate the desired format.

Why the 'cd' driver seems to think 'a' is a good idea is beyond me...

martijn