Video streaming is really bandwidth intensive. This wouldn't be, since the servers only need to tell us what the players look like, the locations where both players and mobs are, and what actions they're doing. Our own client software would put together the actual images based on that just as it normally does.

And having full camera control while watching a battle would be great. We'd actually be able to get a better view than anybody has now, because the players actually fighting in the battle have their Point-of-View determined by where their character needs to be for the sake of performing their attacks and dealing with mechanics. An observer could choose a POV based purely on where gives them the best view.