Issue #17615: Comparing two Unicode strings now uses wmemcmp() when possible

wmemcmp() is twice faster than a dummy loop (342 usec vs 744 usec) on Fedora
18/x86_64, GCC 4.7.2.