It's an interesting technique.
This also seems like it builds off their work with Project Delorean (though it was renamed Outatime due to obvious trademark issues):
http://research.microsoft.com/apps/pubs/default.aspx?id=226843
Same people worked on that in fact.
I do wonder how complicated their predictive models are. For example, if they were doing head/neck modeling in their model, they could likely predict at which point the user would be reaching a point at which it would be uncomfortable to continue turning their head, and if they had brancing prediction like in Outatime, could render multiple predictions at once server-side, ready for multiple options the client could do.
*Edit*
I don't see this working any time soon:
1.- VR requires consistency for Presence not to break which they seem to achieve in their controlled enviroment, but in the real world ISP's reliability, bandwith caps, speed and latency variability all get in the way
2.- They're using an Oculus DK1 which renders at 720p 60fps, the CV1, Vive, Morpheus and even Gear VR render at a higher res and refresh rate, specs that will be the minimum from now on, I'd like to see them test on those and see if they get the same results
3.- DRM, ownership and yadada
It's nice that they're researching this, but by the time any of the above don't get in the way, hardware won't be a problem, at that point I don't even see VR movies using this instead of cheaper specialized mobile chips. The focus should be on rendering techniques and hardware issues, like almost everyone is doing.
Also, I totally expected them to have their own VR prototype by now, it's dissapointing to see them lagging behind because I want VR to be viable in most gaming platforms.
1) Just keep in mind this is still research. If they find this works, but current market conditions would make it unable to release it as-is, they could adapt the technology. For example, instead of rendering on some remote server in a data center, what if you could instead run the server from your desktop machine, and do same-network streaming. Think of it like Steam Streaming, but made for VR. Plus, internet infrastructure is always improving. 5 years from now it will be a different game than today.
2) Agreed. Though, some of your point ties into the above. Bandwidth is the big limiting factor, and of course, the frame rate can depend on latency. However, because they're essentially doing async timewarp, it will be interesting to know how far they could push this. Could they make a good experience at 90 Hz for example?
3) The business model has nothing to do with the technology. See my example for point 1 how they could allow you to run this server off your own machine, using your own game. It's not worth getting into that discussion for a research project.