Hello,
here is the patch [1] which makes workqueue(9) structures more CPU-cache
friendly, as suggested by Andrew, and implements MI CPU ID, which could be
used as an index, as discussed previously. Also, there are fixes for other
discussed problems.
Currently, WQ_PERCPU flag could be used only after all secondary CPUs starts.
For now, it would be OK, but this should be changed in a future. There should
be an alternative for CPU_INFO_ITERATOR and CPU_INFO_FOREACH too,
thought - now it looks weird, from the view of dynamism.

Please review.

[1]. http://www.netbsd.org/~rmind/subr_workqueue.diff

-- 
Best regards,
Mindaugas
www.NetBSD.org