The use of random linear coding of packets is known to be beneficial for both multicast transmission as well as transmission over time-varying communication channels. In this work we consider both scenarios in conjunction and characterize the stable multicast throughput performance. We first consider a scheduling strategy that is able to observe the state of the channel and we provide an upper bound on the throughput. We then consider a random linear coding strategy that performs coding over multiple multicast flows and we provide a lower bound on the throughput. We show that instances exist for which coding outperforms scheduling with channel state information.