Evidence on the Regularisation Properties of Maximum-Entropy Reinforcement Learning