I think you'll be able to listen to casters in game too. There's the capability to send audio streams with the game data too. Similarly camera control can be shown by the casters.
Dota2 implements what HLTV and SourceTV have been able to do, seamlessly.
Basically the way it works is this.
Game is hosted on a server, TV Client/casters connect to that. That master server then can stream out to hundreds of repeating servers. Spectators can connect to those repeating servers. In theory it should work flawlessly and have no impact on the players playing. I wonder if they've deployed a bunch of servers that can automatically allocate more resources to stream popular games. That'd be pretty nifty