It is 100% the cause of the issue. The performance gadget keeps a copy of the last set of PAL stats in memory on the client machine in IE. When we passivate a requestor we persist PAL to perf-stats, serialize/passivate the requestor and remove the requestor from the server completely. When the requestor is activated it starts with new PAL. When the performance gadget requests the PAL it tries to delta the old copy vs. the new and you get negative numbers. The best any bug fix could do here is give you all 0's after passivation, but even that would be a low priority change because this use pattern should never happen.
Thanks for your reply. You meant that a passivation occurs between my PAL readings ?
After each passivation, the readings are reset according to you.
My test duration is about 1 minute length. I'm unlucky if a passivation has occured. I'll continue my investigation.
Thanks for your answer.