Based on my naive understanding of compressed video, it would still send full keyframes every x frames, but the diffs between keyframes should be minimal.

I'd be less worried about that and more worried about the fact that you're using a video compression algorithm on music. That's not optimal!

We have plenty of music streaming services, Youtube seems like an odd choice.

I'm pretty sure that the soundtrack is compressed using some conventional audio compression scheme. Usually, the audio and video are kind of separate streams, which are synchronized by the player using information provided by the container. For example, Matroska video files (*.mkv) are codec independent ( see http://en.wikipedia.org/wiki/Matroska ).

The audio is still compressed using AAC or Vorbis at somewhere around 96-128 kbps depending on format...

Youtube has really good recommendations, doesn't need flash, and has a huge selection and existing community. Worse is better™