For characters animations don't blend too well with each others as some things reset between postures, and the walking animation doesn't translate too well to crowded spaces as people would turn their bodies to maximize the distance between body and objects (up to a point), sometimes even opting to take a different path depending on people.
And the camera is very smooth, which filters out the little balance shifts we all do (even typing something here I notice the subtle changes to vision oscillating in complex patterns which head-bobbing would be a simplified first-order approach to).
I assume it's still very difficult to do all these things with current computing power, and in the case of the camera I'm not sure it will ever be possible for something interactive to feel quite perfect unless you switch to a VR headset and carry your own balance into it.