It also switches to a merkle hash per-file and specifies that each file is aligned to the start of a piece, and the size of the last piece of each file matches the amount of remaining data.
This means in large multi-file torrents you don't have to download (and store) the two extra 1-4mb pieces at the start/end of each file anymore.
The tree structure seems very similar. It would be nice if torrent clients could interact with ipfs or gain ipfs capabilities. Think torrents that update themselves when the files change (thanks to ipns).