Thanks for the new approach, I hadn't tried this one. Unfortunately it's quite a bit slower, and doesn't check for characters in the second string not matched in the first.

leftover: 16 wallclock secs (15.80 usr + 0.00 sys = 15.80 CPU) @ 6481.01/s (n=102400)
l_blakem: 26 wallclock secs (26.07 usr + 0.00 sys = 26.07 CPU) @ 3927.89/s (n=102400)