Try rebuilding a kernel with DEBUG and using sysctl to turn off
clustered reading and writing:

	sysctl -w debug.doclusterread=0
	sysctl -w debug.doclusterwrite=0

I find this cures most problems.  In fact I've hacked my own
kernel source to turn clustered reads off.  That seemed to fix
rather similar problems for me, months ago.

I suspect this is a problem that has wider impact than just the
PMAX port. But perhaps the PMAX-specific SCSI code is less adroit
at handling clustered reads than the `standard` NetBSD code.