> It's about 10% faster, but it's cheating by comparing bytes instead of utf-8 encoded characters.

I'm really glad that you acknowledged this — I work with a lot of non-ASCII text and have run into that more than a few times in real code.