Thought it would be useful for you - to determine two MIDI file similarity

[klesun]

http://stackoverflow.com/questions/6709693/calculating-the-similarity-of-two-lists
http://hetland.org/coding/python/levenshtein.py

Like, you transform all first NoteOn events of a Channel to Integer list (their tune % 12) and pysch-pysch do magic.
Can start with first 20 tunes when searching from thousand of midi-s
Can also compare not first NoteOn tune, but whole Chord (like both has at least one similar Note => equal)