> 
> so I guess all that before is a detailed way of saying:
> we should have a two-level interrupt scheme for the
> high-priority timer device just like we do for the floppy, serial
> and audio devices.

It seems (from just eye-balling the code) that statclock() is not
overly greedy on CPU cycles.  Only when process rescheduling is
required can it induce a context-switch. Maybe it is simpler to
handle this case by lowering the IPL (to 1, or so) just before
entering the AST trap handler.

-pk