Generally comparable perf for the "good" case where memchr doesn't return any collisions (false matches on lower byte) but clearly faster with collisions.