Created attachment 385180 [details]
We're a lot slower than Safari 4 on all three parts of the testcase linked in the url field (and attached, with a doProfile boolean to flip to true if you want to shark it).
We're spending a bunch of time in JS_ComputeThis, xpc_qsXPCOMObjectToJsval. Those two account for over 30% of the runtime for the "node creation" approaches. Also spending a fair amount of time on the final insert; in the innerHTML testcase this is 70+% of total time. We should use this bug as the tracker and file separate bugs on specific issues here, or add as deps bugs that are likely to help (e.g. slimwrapper).
Nightly / Chrome 32 / IE 11
Straight-forward node creation 0.413s / 0.365s / 3.034s
Straight-forward node creation, but no lastChild.lastChild 0.392s / 0.207s / 2.884s
Using innerHTML instead 0.3s / 0.278s / 5.339s