We had a problem where the WMES wiki was painfully slow. The access log showing thousands of Special%3ARunJobs entries (NOT one per request).
$ grep -c Special%3ARunJobs access.log.2014-06-03
$ grep -c -v Special%3ARunJobs access.log.2014-06-03
(that would be 133 RunJobs per request‼)
Changing $wgJobRunRate to 0 from the default of 1 “fixed” the slowness.
From the external appearance, it looks like each Special:RunJobs then called another Special:RunJobs with an infinite recursion until the time limit.
There are 4 jobs with old dates in the queue, but even if they are not completable, they shouldn't have produced so many POSTs.
At the point $wgJobRunRate was disabled, the wiki was running 00f91912c64 (master), but it had been ‘broken’ for weeks.