Issues with blind hashing

It is easy to inadvertently leave clues that would enable an attacker to distinguish real vs. fake hashes or even map users to hashes
Filesystems, databases, web servers, etc. might store meta-information such as timestamps, or/and they might reveal likely relevance between records through use of adjacent filesystem blocks, journal records, or/and log file records
Fake salts and hashes themselves might be inadvertently distinctive
Partial hashes, Bloom filter, or even a bitmap are sufficient to rule out most {user, candidate password} combinations - can save space on attack nodes
e.g., a 1 GB Bloom filter will rule out 98% of possibilities against a 1 billion entry table, whereas a 250 MB bitmap will rule out 60%
on the server, the same 1 billion entry table would be many gigabytes