Variance reduction techniques for stochastic proximal point algorithms