Requeries should use the file hash. Suggestion: for now send two queries, one for the file name and one consisting of the SHA1, and start considering a file's SHA1 to be one of the things that can match a query term, as well as artist, album, file name and directory names, and so on.

When enough clients out there regard a SHA1 of a file equaling a query term as a hit, then you can drop the original query and only send the SHA1 query, reducing network load.

Requeries should be much better then.

Also, zero length files should not be shared. They're automatically useless -- anyone who wants one can create their own by means of right click, "new text file", and rename to suit on Windows and by using trunc on any file on any sane operating system.