Growing demand for multicast commun-ication in large network settings has focused attention on the scalability of reliable multicast protocols. Our paper uses both simulation tools and experiments to compare two scalable protocols, focusing on an aspect not often studied: we emphasize stability of latency distributions as these protocols scale, although also considering overhead and link utilization. These properties are considered in a variety of network topologies and with several levels of packet loss. Our findings confirm that SRM scales poorly under some conditions: to obtain reliability, the protocol incurs overhead linear in group size and throughput fluctuates erratically. We also show that SRM latencies can be very large and that latency distributions are unstable as a function of group size and network topology. Our own protocol, Bimodal Multicast, also exhibits overhead growth, but the rate of growth is slow, and latency distributions and delivery throughput rates are stable.