I don't see any reason why you should not use the arithmetic mean.
If you know the distribution of the measurement, you may wish to use the estimator constructed based on that specific distribution. But first I don't think it's easy to get the distribution. Second, even so, your sample size is small (=3) while the optimality of maximum likelihood estimator is in the asymptotic sense and is only suboptimal for your case.
So a way that is both easy and efficient is the arithmetic mean!