It should be good enough for most MMOGs and also to run AI and even physics of objects far away from the player on the cloud, as it means the cloud could send you refreshed data 20 times per second. But even for far away scenery graphics, though, it would be still eccessive, unless you accept them to be updated at 20fps and admitting the bandwidth is enough, although I guess it should be possible to encode those data on the fly and send them as a compressed video stream, like an mpeg, but I don't know whether the GPU can combine its locally generated graphics at higher fps with a remote video stream in a seamless way along an irregular edge between the two parts: surely every GPU can show a video stream in a squared window inside other graphics generated by itself in 2D or 3D, but seamlessly join a stream to the local graphics along an irregular edge is something I never heard of, also because when you encode those graphics as a video stream you lose every 3D information about the images, so the GPU can do simple things with it, but not having anymore infos about position and depth of those remote polygons, it can't apply 3D operation to them, for example to calculate clipping, and assumptions about locally computed graphics being closer, and cloud computed parts being far away could still leave holes in the frames, unless a very conservative redundancy is applied.

doesn't say anything about other performanceincreasing aspects tho.

tbh he isn't the one working with server connectivity on the scale microsoft is planning to, no wonder he is hoping/saying it isn't working well.... YET

oh yes that is so gonna not work, but secretly wait for the results and hop in if it does lift off? thought so

It may sound weird but i would prefer the graphical quality remains constant at say 80% as opposed to varying between 80% and 100%.

The cloud while full of potential will most likely have a few issues in the beginning with mass implementation.

I don't think MS has ever implied that their cloud compute would be directly rendering graphics. Consider that offloading heavy compute tasks such as AI pathfinding frees up local resources, which indirectly leaves more capacity for local rendering.

Yes, I agree, but since so many people still think it could be used for graphics too, I was just trying to figure out how it could be done and the problems that could arise from all the existing limitations and the compromises these could force to accept. One possible compromise could be to encode (in a lossy and highly compressed way) far away graphics computed by the cloud as a video stream to send them back to the client, but, as I wrote, I suspect that could solve one problem (bandwidth) and create several bigger ones.
About that, Cerny said an obvious thing, but it looks like even repeating it thousands times, and confirmed by techs too and Microsoft itself, isn't enough to persuade some people.