> I sure couldn't take that attitude with the kernel, though
> (it's already causing problems with the sun4c serial ports).

I'd bet long odds that suboptimal code has nothing to do with serial
problems on NetBSD.  *something* is shutting off interrupts for obscene
periods of time.  Something machine-independant, because this problem is
seen with varying frequency on most architectures, including bloody fast
Pentiums that are about 100 times the raw horsepower of the decrepit
68020 box I used to run multiple simultaneous 38400 connections on with
infamous Zilog SCC serial ports.  (Granted, there was a separate 68000
handling front-end processing for the serial ports, and the OS running
on it had a lot of thought put into serial port handling, but raw horsepower
ought to be able to make up for a large number of sins...)