The NFLT mainly applies across all functions. So, I don't see a prima facie reason that there can't be a best overall algorithm for the problems we tend to encounter, since these are an infinitely small subset of all possible problems and likely share common characteristics.

In fact, to be a total contrarian, I suspect intelligent design theory can provide such a heuristic for all real world problem domains. The irony is thick indeed.

> Offtopic : I am frustrated with people ignoring no free lunch theorem and trying to use GAs everywhere.

GAs don't violate No Free Lunch. After all, you're a genetic algorithm...