Google Scholar has picked up on Krishna Palem's previous publications. There should be a law that internet articles about scientific research have to back everything up with links to a journal article or something.
The power consumption angle is quite interesting - but I given the power saving they demonstrated wasn't really of an order of magnitude or more, then it might be hard to make it scale up to practical tasks and have errors worked out of it. Perhaps there are more savings to be made, but I would expect to see orders of magnitude less power for this stuff to be practical.
The libavcodec has an option that does exactly that. It skips steps in the filter loop that only affect the single frame. It makes things quite a bit faster.
http://scholar.google.com/scholar?q=Krishna+Palem+probabilis...
For example, here's an interesting paper that talks about saving power by making AND/NOT gates probabilistic.
http://www2.computer.org/portal/web/csdl/doi/10.1109/TC.2005...
I guess there's an inherent assumption that that reduced power consumption will scale up to the level of "checking email and browsing the internet".