[hcks]

In the 1D case the single point will be the best estimator for the mean no matter what the variance is

[pfortuny]

OK, this is a good reply, and very informative, many thanks. Now I understand why the thing is paradoxical.