Maybe you are asking not only about remote execution, but also mobility of code. This is a problem that is previous to mapReduce, since mapReduce assumes that all the code (and the data) is in place in the respective nodes. In fact, the distribution of resources in order to efficiently use mapReduce is a design problem that the google people has done by hand.
But my intuition says that there are a general algorithm for distribution of code, data, bandwidth and resources in general that moves around at execution time to achieve better and better performance for a given grid of nodes and for any task, for example, a mapReduce task. I would be very interesting to read something about this.
I know that some efforts have been carried out the past , for example mobile haskell