I'm proposing again my idea of a Java-based client-side search engine:
http://quatramaran.ens.fr/~monniaux/wikipedia_en/search_applet.html
(Need good JVM, NOT the old Microsoft VM.)
The index is downloaded by chunks, as needed by the client. There's some
caching.
[The index was generated 6 months ago from a database dumb. I'll remake
it once I get mysql working again.]
What do you think? If interested, I may implement full-text search.