Yes, I mean performance of Shark itself. It's been annoying me no end that Shark is so slow to actually produce the profile, with lots of time spent by pilotfish (presumably gathering symbols). I've experimented with this some, but running dsymutil and then stripping didn't seem to help much.
Eventually, I discovered that simply running
strip -S on my XUL library (in an
--enable-libxul build, of course) made Shark produce the profile quickly. I can't get source code display, but I _can_ get the various callstack information, which is what I usually want. And getting the results in seconds instead of minutes is absolutely worth it.