>On the same topic we have kern/1515, which could be fixed at the same time.
>The issue here was that on slow machine the perf hit could be noticable,
>but no definitive statement was made (at last I didn't find one in my
>archives).

I could be wrong, but i think the consensus was `do it'.  If a machine
is really slow (vax 11/750 the packet rate it can handle is low enough
that two extra loads and stores and adds-with-carry arent going to be
a killer.

That's what I recall the m68k folks saying, anyway.