Rather thin on the ground in expertise if this sekerob is needed. Searching on this member and a few key words pops up this post though: http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=440196 . Looks like any task that has a remaining deadline of less than two times the actual buffer causes such a task to skip ahead.

Boinc 6.10.58 often did not start resend workunits soon enough and would go into crisis mode trying to get them done by starting and stopping partially completed workunits in a seemingly ad hoc manner. To prevent this from happening we went to great lengths to start resends as soon as possible. One major improvement in the newer versions of Boinc is that it is much more intelligent in how it schedules workunits. It rarely (if ever) goes into crisis mode, and resends eventually get crunched without pausing partially completed workunits. There really is no reason anymore to try to get resends started immediately.

Well said. The only time I have seen BOINC go into crisis mode(high priority) is when the task lengths have become highly variable on the upper end (task(s) probably 2 or more times greater than average), which throws off the initial BOINC calculations. If left to its own devices, BOINC will gradually return to a more normal mode. The ability of the program to get all existing jobs completed in a timely manner works remarkably well IMHO.
Cheers

That change was made in 7.2.36 as noted here.