The range for the slowdown is http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=46bd519c8e2b&tochange=0be44e8bf22a which includes PJS removal bugs, bug 1117753, bug 1117735, bug 1117764.

If the PJS removal is the source of the regression then the regression is almost certainly caused by perturbations in object code for the GC, since a bunch of GC code that was previously active on Nightly - in the form of checks for whether the nursery was the ForkJoin nursery, and some functions defined on multiple nursery types - no longer is. In particular I could imagine some functions that were previously not inlinable due to size or complexity becoming inlinable after these patches landed.

Looks like the regression went back down with the landing of bug 1118038. It does sound like some kind of minor perturbations, I guess we can close this.