Well, I have thought about that problem (fake files) too..
If *they* (no idea who I mean) think a bit further *they* could code a client which sends queryhits to all queries with some better filenames like:
Query = gescheiterten existenzen vogel
Queryhit = "Gescheiterten_Existenzen - Vogel.mp3

You just have to set up a database of the artist or program name to do so.

Then *they* could just let *their* servent send 000000s
till the size of the file is reached. (Usually a mp3 song is about 4 mb big, so send a file with size = 4 mb + - Random value which has only 0s in that file.

Hashing wouldnt stop that either because the hash is different if you add some more bytes to that file, or am I wrong?