I've noticed quite a few people asking about this around the internet, and the general consensus seemed to be that it wasn't possible.
Turns out it is - I was looking into this for a recent personal project and determined that it is indeed possible to make MTAudioProcessingTap work with remote streams. The trick is to KVObserve the status of the AVPlayerItem; when it's ready to play, you can safely retrieve the underlying AVAssetTrack and set an AudioMix on it.
I did a basic writeup with some (mostly working) code here: http://venodesigns.net/2014/01/08/recording-live-audio-streams-on-ios/
If you already managed to handle this, more power to you, but I figured I'd answer this question since it comes up pretty quickly in Google for this stuff.