In cleaning up the detritus left behind in my local task log database from the last meltdown, I noticed that one of the tasks which I had received from just before the project ground to a halt got a 221 Redundant Result abort from the project when I reenabled MW on my hosts.

Experience over on SAH with the workload burn rate and host volume they carry showed they could not handle having the advanced features like Lost Work Autoresends or Redundant Result Aborting enabled without causing the serious negative effects on DB performance in other areas periodically.

So it might be worthwhile to see if disabling 221 functionality helps, since it's pretty clear now that the periodic completed task purge daemon isn't the primary factor causing the problem. It's far more likely that having to search every host's whole DB record every time there's a scheduler request to see if any work in progress is redundant is the culprit.

Alinator