When you start talking about latency, you are inherently talking about Amdahl's Law because the question that gets begged is, "can we make amount of work X complete in less time with more processing power Y" ? And in your problem domain of video coding, esp if it is for real-time face to face communication or telematics, latency matters. So Amdahl's Law matters because it dictates the bound on latency.

Doing more work X (making the pie bigger) in about the same time as Y is http://en.wikipedia.org/wiki/Gustafson%27s_Law, latency, like the time of light it is one of those physical universal properties we won't be able to tech our way out of. Latency is inherently a serial problem. What changes when we could try every solution at the same time?

Nearly all hard problems we are solving right now don't need to have exact solution. They can be time bounded or quality bounded or both. What we lack is a clarity in being able to create algorithms that are probabilistic and progressive. Algorithms that refine and converge towards a solution over time. Better languages can help with that.

Here are some examples of probabilistic parallel minimax algorithms.

http://www.top-5000.nl/ps/Parallel%20randomized%20best-first... http://www.math.md/files/basm/y2010-n1/y2010-n1-(pp33-46).pd...