Any reason for using IO::Compress::Zip instead of Archive::Zip? It allows you to specify the internal path and file name (Unix format) when you add a member to the archive. Just be sure that your path and filename are unique though, it does allow adding multiple files with identical filenames.

I have done this before using Archive::Zip. You can create the zip file in memory, read in your files, modify them in memory, add them to the zip from memory, and then output via HTTP without having to write anything back to disk (assuming it all fits).