First hypothesis would be be b2g process doing too much work + process priorities hurting things, so bug 761933 is first to check. If the Marketplace is loading a significant amount of resources from the network, as it appears to, then we'll be routing HTTP buffers in the b2g main thread at a higher priority than we parse+execute+reflow+paint them in the child, which is wrong. Possibly orthogonally, sticking the profiler on this may show low-hanging fruit.

Adding direct dependency on bug 784647. We do layout/gfx-y things off of XPCOM events, but HTTP data is delivered through IPC. Prioritizing IPC is likely artificially deferring layout/gfx, which is bad for user-perceived perf.

I noticed a qualitative improvement after bug 784647. There may be more to do here, but let's do it elsewhere.