Hacking Into Pandora's Box

I have several current hacking projects and one of them that I keep coming back to is Pandora , the internet radio service. You tell it a song or artist you like and the service customizes a radio station just for you with music which has the same qualities as songs you like. They don’t offer mp3 downloads however, and they don’t allow you to fast forward, rewind, or skip songs. It’s a flash-based web interface, and my digging (both on windows and in linux) has led me to realize that these songs are temporarily being downloaded to the hard drive. (They’re in M4A format, which is essentially videoless MP4). The problem is that these files have garbage filenames (random letters). If I can just obtain the artist and song title to associate with each, I’ll be able to write a Python script which will allow me to copy these files as they download, assigning their ID3 tags accordingly.
I know this is possible, because I can spy XML data being transferred over my network (thanks to tcpdump) which contains all this information. The timing corresponds to the download of each subsequent M4A file. Where does this XML data go? This is the tough question. I could accept the fact that it is loaded directly into flash and begin attacking it at the TCP protocol level, but the fact is that a lot of weird files show up in /tmp which I have a hunch are these XML files. The problem is that when I go to open them, they’re blank. My theory? Pandora is downloading files and keeping them only as long as they’re needed. The M4A files are readable for as long as the song plays, then they’re deleted. Perhaps XML files are being downloaded and deleted after they’re needed (1/100 of a second later), so they’re hard to “catch”. Alternatively, I see a lot of “temp.js” files which are empty. Perhaps they’re being overwritten with /dev/null rather than being deleted…
My current approach is to write a Python script to log file activity (creations, deletions, modifications, etc.) which I hope can be run at least 1,000 times a second (am I dreaming?) to elucidate this issue. Optimal scenario: I’m able to identify XML files as they make their brief appearance on my hard drive, copy their contents, and use this data to attach meaningful names to meaningless M4A files on my hard drive.
Below, a photo of me in my quest. (I’m working on my new laptop in the confocal microscope room as tissue sections are being scanned into the computer) Yay linux. Yay python.