Try to figure out how to split up huge indices to avoid timeouts
Multiple pages, transcluded onto one?
Probably need to have multiple actual pages - otherwise it's just unmanagable. Will probably need new target syntax. split=alpha, segments=A-C,D-G,H-L,etc?
Address scalability of cache (filename hashing or something)
Make cache more efficient by caching compiled objects (%index hash bits)
Allow indexing of talk page if opt-in is on a different page (like in a transcluded header)
