A review of using the Differential Evolution algorithm to fit reflectivity data has been published in J ournal of Applied Crystallography [Bjorck11]. If you do not have access to the journal just send me an e-mail email@example.com.
In summary one can say that choosing a minimum population size of 30-50 and k_r = k_m = 0.5-0.9 provides a rather robust algorithm for reflectivity refinements. If you need more stability the population size can be increased and/or the k_r be lowered. For more details see the paper.