Tp appears to have regressed on birch sometime on 11/28 or 11/29. Noise in the data makes it hard to pin down an exact changeset where the regression occurred: http://graphs-new.mozilla.org/graph.html#tests=[[84,27,20]]&sel=none&displayrange=7&datatype=running This might be caused by the viewport code from bug 694901 changing the browser size more often than necessary.
Created attachment 577786 [details] [diff] [review] instrumentation patch This patch adds some timestamp logging to the viewport code; it doesn't identify any noticeably expensive operations during page load. An alternate theory is that we recently fixed bugs that caused us to paint only part of the page; now that we are painting the whole page, rendering might be more expensive.
After running some more Tp4m benchmarks on the surrounding changesets, it's clear this is a regression from bug 694901.
Matt, this is a P2 for you to make a call whether this can be fixed or we can live with the regression
This seems to have been caused by an increase in the number of redraws and size changes, which was fixed by work like bug 709492.