I have a few things to note about this patch:

1) It would be less of a performance lose to simply make p_priority
and p_nice be ints; they wouldn't require extra locking when written.

2) It's not necessary to call splstatclock() when reading a snapshot
of p_nice, p_priority, or p_estcpu (such as in retrunnable() or
proc_compare()), providing they're marked as volatile.  You're not
even consistent about this anyway.

3) You made exactly the mistake that the big comment in userret(),
right above the code you modified, mentions.