jsPerf is down today, but this article suggests that the performance gains from caching jQuery selectors would be minimal.
This may simply be down to browser caching. The selector tested was only a single id. More tests should be done for more complicated selectors and different page structures...