I feel like a cat watching squirrels on the other side of a window; stories have been posted about genetic programming and Java on microcontrollers, yet I can't comment on them.

One article started with the assertation that programs derived via genetic programming are highly efficient. This is, for the most part, flatly incorrect. Genetically-derived programs are typically bafflingly complex with all sorts of redundancy and indirect mechanisms that would make Rube Goldberg proud. What genetic programming is good for is producing solutions to problems that you don't know how to solve yourself, but can tell a correct solution from an incorrect one.

Although, I suppose if program run-time were one of the factors considered in evaluating a program's 'fitness,' you could evolve for efficiency. I don't know what that would do to the number of iterations required to find a solution, however.