I have a 340 with a 98658A.  It runs rather happily with a whole stack of
Conner CP3040As with Apple ROMs that I acquired cheaply (The local CC was
taking them out of Macintoshes by the hundreds).  I run them with one root
disk, and six disks acting as a ccd (concatenated mode).  They run great,
much longer than any of my i386 boxen.  Unfortunately, they are still seven
disks, and for practical purposes, they generate a fair amount of heat, and
I've yet to build a case for them.  (They stand on end, all bolted to a
three interbolted pieces of cheap bracket)

I have at my disposal, a DEC RZ23 (about a 100M disk), a Quantum 105 (LPS?),
a Quantum LPS40, eight of the 40M Conners, and a SCSI-1 cdrom drive which 
claims to be a Matsushita Kotobuki CDR something or other.

My quandary is this:  After running for some short period of time, some
times long enough to mount and fsck, sometimes long enough to tar all of
/usr over from an nfs mounted partition, the scsi driver reports an abort
from target 0 (the disk in question):

scsi0: targ 0 abort from go: phase=0x8a, ssts=0xb1, ints=0x0

This will happen with both the Quantums, and the CD-ROM drive.  The DEC disk
gets reported only as an erroneous st device, possibly because it's never
sent the spinup command.  If I power down the disk, I can work with it for a
short time more before it occurs again.

What I would like to do is work from one disk, preferably in the 100-300M
range.  If I'm missing something basic, please let me know.  If you are
using a SCSI disk with a similar setup, please let me know.

I'm running (and have been) a 1.1 GENERIC kernel, and Jason's latest
snapshot generic kernel.