Admissibility of Completely Randomized Trials: A Large-Deviation Approach