I wish I had some better inmup for you than what I'm about to offer, but first off, just about anthing is possible! 🙂 I like your ideas and think they're probably the most interesting I've seen relating to the particular subject of "monitoring" and http upload.
Anyway, as far as "predicting" the next temp file name in UNIX, I wouldn't know where to start. However, your idea of scanning the directroy for the most recently updated file would do the trick.
For a given user that starts an upload, IMMEDIATELY "associate" the most recently updated file to him.
Track that file's size, thus giving you a basis for a progress monitor.
Yes.. I think that would work. Keep us posted on how the project develops. Good luck!