I agree on this -- I wrote my thesis on the VRP (Vehicle Routing Problem) which is an even more difficult problem. I've studied GAs extensively (they are pretty awesome, hence my interest), but other methods are much better. I've heard great things about ACOs, but eventually I settled for Tabu Search, because:

- It is deterministic, hence not a black box - It is a local search, easy to explain - Does not require as many parameters to tweak (GAs' performance demands on how you set these: mutation rate, population size, cross-over mechanism (dozens of possibilities), stopping procedure, selection procedure, etc...) - Considered by literature to be one of the strongest class of algorithms for the class of TSP problems, in both solution quality and computational effort

Here's an open-source implementation in Common Lisp: https://github.com/mck-/Open-VRP