I love Pratt parsers (or Top Down Operator Precedence parsers rather), I recently re-wrote the IronJS lexer and parser by hand (from using one generated by ANTLR) using the techniques described in this blog post and saw some pretty hefty performance increments http://ironjs.wordpress.com/2011/03/19/new-lexer-and-parser-...

Edit, I also have a generalized version that can take pretty much any input and any output (F#) here: https://github.com/fholm/Vaughan