Yes of course, the mean is the first moment and a Bessel correction would be inappropriate. Now I feel stupid. The mean calculated by sum(p_t x_i) or 1/n sum(x_i) is already the best linear unbiased estimate. Maybe we can't get better than twice the mean?