Since the writing of StringImpl::computeHash, the hash function used and sited, Paul Hsieh's SuperFastHash http://www.azillionmonkeys.com/qed/hash.html has been updated. We should investigate whether there is any win to using the new function. Maciej suggests that looking at both performance and probe chain length statistics (found using the dump statistics mode) are important.
<rdar://problem/5708284>