Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods?