In case the current blob index becomes too much overhead:
sparkey has a similar concept to the blob index (two files, the index one can be cached more), only with binary hash index lookup.