according to Binmore (http://jasss.soc.surrey.ac.uk/1/1/review1.html) Axelrod didn't actually run his simulations for long enough - when they're repeated w/ a longer time horizon, tit-for-tat proves not to be the ideal strategy.
I think Axelrod actually claimed loudly that tit-for-tat wasn't optimal, and offered a few other strategies that would have done better (including tit-for-two-tats, interestingly enough). He was also pretty careful to add that a lot depends on context and that the aspects that tit-for-tat embodies (being nice, reciprocal/retributive, forgiving, and clear) are what's really important.
that's true, but I think (could be wrong) that Axelrod presented his results as "definitive" given that no-one had simulated a comparable timescale before. it was certainly presented as such in the textbook I was reading from.
