Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?