> This brings up a good point - should we work towards establishing some QA
> benchmarks that we measure against whenever we feel we've hit a stable
> point, beyond straight performance marks?

Well, we've got a decent criteria list:

1) mc0 performance > 10k/sec
2) serial port does not lock
3) devices behind bandit properly share interrupt

Separately,

4) mesh works at 10 MB/sec
5) multiprocessor machines don't lock

That's a decent list!

John