I've done vector indexes on CouchDB before, because it supports arrays as keys in it's map-reduce implementation. Worked great for fast document similarity search.

Brief writeup: http://splatcollision.com/page/fast-vector-similarity-querie...