The latest issue of Communications of the ACM includes an article (The Reality of Simulated Actors) that estimates that it will be 20 years of computing power advances before we'll see digital actors truly capable of verisimilitude. 20 years == 4 orders of magnitude. One can't help but wonder if grid computing couldn't bring this to fruition significantly faster. What would be involved in creating Olivier @ Home ?
The author, Alvy Ray Smith, divides the problem into two parts -- the "model problem" (representing the appearance of reality in a convincing way) and the "control problem" (the interface to the model). He sees both as requiring major increases in computing power, but more importantly, the "control problem" may be "difficult" (e.g., you can't hand-animate all the surfaces involved in an amused glance).