Hello Richard,
The bug was already traced by me to "sched: fix fair sleepers" (which you
list as well), and this was reverted just before the final 2.6.25 release.
Can you confirm that the issue is fixed for you as well with v2.6.25?
Unfortunately most of the communication about this was done by private
mails, although I did update the bugzilla entry and replied to the
"regressions" mails from Rafael. It's a pity if this has caused you
unnecessary work.
Note that Ingo has just resubmitted the (probably same) patch for 2.6.26
[1], so we're not done with this yet.
Cheers,
FJP
[1] http://lkml.org/lkml/2008/4/19/166 (see also my reply)
--