Bugzilla

I worked on this a bit.  I updated the patch (only one needed now): [patch](https://hg.mozilla.org/try/rev/5424c3a2b144da3ca1d9ea7f61e8782fd518f18f)

I ran Talos and the results weren't great: [Try Run](https://treeherder.mozilla.org/#/jobs?repo=try&revision=e9b817f8a37addc43a3c5fff144eeacaa7f45f05) / [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=e9b817f8a37addc43a3c5fff144eeacaa7f45f05&selectedTimeRange=172800)
Some 2-3% regressions.  And yet, you'll find this was actually the best try run I had...

Then I tried to only apply the separation to non-System Principal: [patch](https://hg.mozilla.org/try/rev/6d3efbc00fa77636b4e6ef73007c57e6d02ae019)
Talos for that was significantly worse: [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=2f1bef963074c1698702eb412fa7e463e8ed6dba&selectedTimeRange=172800)

Then I noticed that previously the data was able to be background-finalized but we had lost that capability, so I added it back: [patch](https://hg.mozilla.org/try/rev/5750573fb7a49ca608e89b6fdb9251af8616361d)
That (plus the system principal patch) also did poorly on [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=dc9f189eb6939b62bf0e5c91c47fe7ab874d300f&selectedTimeRange=172800)

And then I ran it with background finalization but without the system principal exception: [Try](https://treeherder.mozilla.org/#/jobs?repo=try&revision=8820c893ad2d646027ebdccf675a03bd317e99d7) [Talos](https://treeherder.mozilla.org/perf.html#/comparechooser?newProject=try&newRevision=8820c893ad2d646027ebdccf675a03bd317e99d7)
I was able to get a really nice win on tabswitch on windows (20%) but a bunch of regressions elsewhere - again more than the very first try run.

I worked on this a bit.  I updated the patch (only one needed now): [patch](https://hg.mozilla.org/try/rev/5424c3a2b144da3ca1d9ea7f61e8782fd518f18f)

I ran Talos and the results weren't great: [Try Run](https://treeherder.mozilla.org/#/jobs?repo=try&revision=e9b817f8a37addc43a3c5fff144eeacaa7f45f05) / [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=e9b817f8a37addc43a3c5fff144eeacaa7f45f05&selectedTimeRange=172800)
Some 2-3% regressions.  And yet, you'll find this was actually the best try run I had...

Then I tried to only apply the separation to non-System Principal: [patch](https://hg.mozilla.org/try/rev/6d3efbc00fa77636b4e6ef73007c57e6d02ae019)
Talos for that was significantly worse: [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=2f1bef963074c1698702eb412fa7e463e8ed6dba&selectedTimeRange=172800)

Then I noticed that previously the data was able to be background-finalized but we had lost that capability, so I added it back: [patch](https://hg.mozilla.org/try/rev/5750573fb7a49ca608e89b6fdb9251af8616361d)
That (plus the system principal patch) also did poorly on [Talos](https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&newProject=try&newRevision=dc9f189eb6939b62bf0e5c91c47fe7ab874d300f&selectedTimeRange=172800)

And then I ran it with background finalization but without the system principal exception: [Try](https://treeherder.mozilla.org/#/jobs?repo=try&revision=8820c893ad2d646027ebdccf675a03bd317e99d7) [Talos](https://treeherder.mozilla.org/perf.html#/comparechooser?newProject=try&newRevision=8820c893ad2d646027ebdccf675a03bd317e99d7)
I was able to get a really nice win on tabswitch on windows (20%) but a bunch of regressions elsewhere - again more than the very first try run.

Next I'm going to try with fewer AllocKinds as suggested.