CGI is great for creating photorealistic things it would be unfeasible to do in live action, like the apes in RotPotA. But I don't see the point in using CGI to duplicate human beings. To paraphrase Isaac Asimov (who was actually talking about robots), if CGI turned out exactly like human beings, it would be a terrible waste -- we've got human beings. The best way to make a movie about human beings is to put actual human beings in front of a camera.
I think if you're pursuing something like creating digital "stunt doubles" for live actors so you can have them do things it would be too expensive, difficult, or dangerous to do with live actors or animatronics, then the uncanny valley is a legitimate concern. But if you're creating a full-on CGI movie -- something which is essentially an animated cartoon -- then it's rather self-defeating to try to make it look like live action. You shouldn't even be trying to cross the uncanny valley in that case; you should just be trying to design and execute the best 3D cartoon characters you can. And for that, I think hand animation is better than performance capture, because part of what makes cartoons work is caricature, the distillation and exaggeration of what's important and the smoothing over of extraneous detail. Beyond that, animators can make their characters "perform" in ways live actors can't, like squashing and stretching the body, having the hair or clothes move in a certain way, playing fast and loose with how gravity or impact affects a body, etc.
Of course, most motion-capture performances in movies are supplemented by animators tweaking the details of the performance by hand anyway, although I think that's less the case as the facial performance-capture technology improves.
