Assigning to chrisha for initial triage.

These heuristics are very primitive, and not able to finely tuned. From observing machines in a lab it becomes pretty apparent that the memory manager tries to maintain a constant size free list, and swap aggressively to do so. We chose these thresholds based on the bulk of those observations.

We know that the threshold is too conservative in some cases, but we don't understand the entire logic of the memory manager, and exactly how it calculates a target free list.

Two things are happening here that will address this:

- We're trying to build a better heuristic based on observation of actual swapping conditions on an individual machine.
- We're making discarding much more aggressive and common place, making it far less likely that swapping will occur in the first place.

Thus this specific bug is a WontFix. We won't be investing any time trying to tune what is a known broken heuristic.