> At that point the devices attached to the isp0 can be accessed intermittenly
> without errors (for instance, there might be a pause of up to 10 or
> so seconds while doing an 'ls' on a filesystem attached to the isp0

Actually, using 'time' I've seen anywhere between 3 seconds real to 
up to 105 seconds.

Why does the system hang for so long, then seemingly recover only to
hang again a minute or (random time) later without logging an error?

I don't mind the stray interrupt messages, but it seems after 4 when
it prints that 'logging stopped' the system hangs like this.  Rebooting
will make it usuable again, but with any amount of i/o on the scsi drives
you are bound to get the stray interrupt messages again.

Is there a way to just let it complain?

I get the same results with the generic 1.6.1rc2 kernel and my stripped
down machine specific kernel (using COPTS="-O2 -mbwx" makeoptions)