A hashing algorithm specifically for passwords that would hash to the same thing based on inverted caps lock or other deterministic user errors (white space at the edges, maybe)? That sounds like a great idea.