Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios