Currently we require users to have a single machine that caches as much of Wikipedia's structure as possible into memory. This is starting to get untenable for larger wikipedia dumps (e.g. enwiki)