Better maxtimedout logic
Imported issue: Initially reported by Alexander Kruppa in https://gforge.inria.fr/tracker/?group_id=2065&aid=16580
Currently, maxtimedout with a default of 100 means that after 100 workunits have been cancelled due to time-out, the server should shut down, because it might indicate a problem with the clients. This is way too simplistic. For example, when using besteffort submission, it is not unusual at all that hundreds of workunits time out, but in this case, the server should keep running.
A better criterion for when the server should conclude that there is a problem with clients will be needed.