It's still missing something. It's just off enough to distract you. But if you turn the sound off, it works much better. I think that's the biggest problem with CG people; it's not their look, it's that we as humans are so accustomed to exactly how the mouth moves that when it's off by even a fraction of a second, we catch it. Better lighting would probably make Kaya more believable, too.
What's weird is that this video was made in 2001, but it's still pretty state-of-the art. Is the progress curve slowing?