Learning in Games: Robustness of Fast Convergence