It would probably be better to build a HoA keyed by file size ((stat ($file))[7]) rather than MD5 sum, with values being arrays of files of a particular size. Any two files of different size cannot be duplicates, obviously. Any hash element that contained just one file could then be discarded, thus avoiding the expense of MD5 sums or file comparisons for a proportion of the files you are testing.

Once you have sets of files the same size you can compare them either by generating MD5 sums, by reading the files (slurping if small or in chunks if large) and doing string comparisons or by using external commands like cmp. (I would recommend against using external commands.) You can save a lot of time by avoiding re-doing comparisons when you have several files of the same size. For example, given fileA to fileE, you would logically start by comparing fileA to the other four in turn, then fileB to fileC, fileD and fileE, and so on. If fileA differs from fileB but is the same as fileE you can see that it is not necessary to compare fileB with fileE because you already know they differ.

I hope these thoughts are of use.

Cheers,

JohnGG