Hi Michael,
 
>   I've still got Greg's diffs between an older version of the asc driver
> that worked, and the one that was failing when I was working with him on
> these problems.  I guess it's time to dig out those changes and try to 
> determine exactly what had changed and how it might affect this particular
> problem.  Since I'm about to leave for 2 weeks, I probably won't be able
> to even look at this until the 2nd week in January.

This would be really fine.
Maybe the following information can be of any help:

The last working kernel I was able to make on my 5000/200 was made on 
May 18th; and the first "bad" kernel after this is from May 21st.
I supped kernel and userland on May 16th 14:24 CET (I use sup.de.netbsd.org,
which maybe has some minor delay) and then again on May 21st 14:53 CET.

Grabbing trough the list of files transferred the only relevant files seem
to be sys/arch/pmax/dev/sii.c and sys/arch/mips/include/trap.h.
Both had changed in the May 21st sup.

Anyway it is my impression that I had a even better kernel several months
before.
The kernel made on May 18th gives me messages like

 /netbsd: asc_intr: data overrun: buflen 4096 dmalen 4096 tc 4040 fifo 15

about once a week during heavy disk activity and then reboots.
But this I think is already a different story.

Merry Christmas and a happy new year to all NetBSD people around the world,

Andreas