Improving Efficiency of Sampling-based Motion Planning via Message-Passing Monte Carlo