Do Reasoning Models Show Better Verbalized Calibration?