Based on the characteristics of packet forwarding in application layer multicast (ALM), this paper proposes a high performance multicast mechanism, in which a NIC-based multicast mechanism is used to send multiple replicas of an ALM packet to different destinations with less CPU intermediation in comparison to the traditional host-based multicast mechanism. By adopting the mechanism, the CPU overhead and the number of times data packets are copied from host memory to NIC buffer. Theoretical analysis indicates that the new mechanism can reduce the delay and improve the forwarding throughput in application multicast. We have modified the NIC driver and added some corresponding APIs to network protocol stack in order to implement the proposed mechanism. Experimental results also indicate the feasibility and validity of this mechanism.