You don't need a stream of video from a device like the Kinect. You can send snapshots of which accountholders are within the Kinect's detection range, their position and attitude in relative space, heart rate and possible key words or terms or even objects, if one was so inclined.
If you can remotely update the logic used by Kinect to interpret its audio/visual data, you can add new terms or objects that Kinect can seek, and can have the device return small metadata related to any hits on that search criteria.
In short, when you have a device that's a freakin tricorder with an internet connection that is always on, the possibilities are almost endless, and all without a perpetual stream of audio or video.
The future of gaming is here? Holy crap at the potential...