This is SiteSearch v0.1. It's fairly efficient compared to scripts that walk your whole HTML collection for each search. So the algorithm is good. That's the hard part. But the implementation is not optimized. (No holding the whole index in memory for all passes.) But that should be just fine for 99% of the web sites out there. If you really cared about speed all that much, you'd write a servlet anyway, now wouldn't you? ;-)
My home page, which contains contact info, etc, is http://www.etla.net/~willey/