Yup, that is exactly what I am saying
64 cases of it in fact in the large library.
I'll try multiple hashes. I wonder what is faster - reading a larger number of files once and computing two hashes, or having the third pass work on the smaller subset but having to read the files again...