i noticed that when the plan 9 scheduler changed quite
some time ago, inferno's interactive
responsiveness degraded compared
to Windows, where it's quite good;
under Plan 9 it was still much better
than it was under any Linux variant
i've tried.  the recent plan 9 scheduler changes
improved it, but it's still not as good as it once was.
(we've got a fairly objective test for it, so it's a little
more than just an impression.)  i did change
the way some things were done to evade faults
in the underlying scheduling (eg, sched_yield can be unhelpful),
but something still sticks on some host systems.  that's also when i
discovered just how appalling Linux's scheduler actually is, and has
been years.  good source for a good sneer, anyhow.
as well as brucee's suggestion about graphics, it might
also be different scheduling in MacOSX.
Inferno's unusual on most host systems in using
possibly many host processes in a single application.
that's when you start discovering the (possibly missing)
fine print on the various `threads'/processes manual pages...
and the implementations.
Plan 9 is fairly good with rfork, to pull this back nearer list topic.
let me know if you find a good (ie, realistic) test and profile it.