On the limits of the hadronic energy resolution of calorimeters