I would say this depends on the browser. Nothing beats an experiment, in this case a simple JavaScript would probably do.
Update: I meant to do something like this:
CSS Speed Test
…but the time differences caused by different selectors appear to be too small to measure, just as annakata said.