Starting around November, the noise band for the tp5o_shutdown_paint shutdown-time test has increased manifold, almost to the point of being meaningless. http://graphs.mozilla.org/graph.html#tests=[[273,94,25]]&sel=none&displayrange=365&datatype=running
Looks like this was discussed back in November but never acted on(?) https://groups.google.com/forum/#!msg/mozilla.dev.tree-management/r9iehVDynmM/zKdgu0lhkeAJ
FWIW, I've noticed that recently (though probably later than November) shotdown times seems quite random, _sometimes_ with addon-manager related errors on the console during shotdown. I don't know if these are related, however.
this became a problem when we switch our process management from the old bcontroller.py method to using mozprocess. The motivating factor there was to fix a large common crash in dromaeo on windows- we did that and haven't had any problems since, but this shutdown stuff is a fall out from it :(
we don't run this test anymore, if we do wish to do it, I would prefer a new bug be opened up.