Tweak the 'load balancing' algorithm
In order to reduce contention we randomly skipped jobs, but this
caused many jobs to end up stopping early. Now instead we keep on
going until we time out (also increased the chance of doing work).
Patch by: Sverre Rabbelier
<html>
<head><title>We ran out of time</title></head>
<body>
We ran out of processing time while serving your request, please try again.
</body>
</html>