In all seriousness, is this problem widespread enough that we should make an
   attempt to quantify the successes and failures of certain machine/disk
   cominations?

That's rather a sad way to attack things.  It seems so far people have
given anecdotal evidence that

	- certain features fail
	- error recovery fails
	- the problem is most likely NetBSD kernel timing (works with
	  VGA console, fails with 9600 baud printfs to serial console)

Which is to emphasize that certain disk/controller combinations work
but _only under certain circumstances of timing_ caused by the platform
and what else is on the platform at the time.  That seems to dictate
finding the timing problems is a better solution, since your catalog
will either not be reliable or will have to give entire equipment
lists.  And you want error recovery in all cases, I would think.

A better page to start might be a "lore" page on how to maintain these
drivers, so people with spare cycles don't have to start by writing for
manuals and begging for details.