We propose an optimization algorithm for joint relay selection and source and relay power allocation under mixed line-of-sight (LoS) and non-LoS path scenarios for both power saving and robustness enhancement of cooperative multicast in millimeter-wave wireless personal area networks. Our aims are to reduce power consumption and enhance the robustness of cooperative multicasts in millimeter-wave wireless personal area networks. First, we describe a novel beam training protocol that is capable of overhearing and information feedback to filter relay candidates with non-LoS links and avoid selecting relays for transceivers with LoS paths. Second, the joint relay selection and power allocation issue is formulated as an optimization problem with the objective of minimizing the maximum combined power consumption of the source and relay under maximum tolerable outage probabilities and transmit powers. By introducing relaxation and Lagrange multiplier methods, a closed-form expression for the joint relay selection and power allocation is obtained. Finally, simulation results indicate significant improvements in terms of both outage probability and power consumption over the conventional combined transmit power minimization algorithm.