Controlling the reinforcement in Bayesian nonparametric mixture models