We present a delay analysis of one sender- and two receiver-initiated classes of reliable multicast protocols. Besides the average delivery delay we consider the delay to reliably deliver all packets and the round trip delay. The former two examines the delay between generation of a packet at the sender and correct reception at a randomly chosen receiver or all receivers, respectively. The latter is the delay between generation of a packet at the sender and reception of all acknowledgment packets at the sender. Our numerical results show that receiver-initiated protocols provide significantly better scalability for large receiver groups and transmission rates compared to the sender-initiated protocol. However, the delay of the sender-initiated protocol within its scalability range is substantially lower compared to the receiver-initiated protocols. To assess the quality of our analytical model we have compared the analytical results with a SRM-like protocol simulation. We show that the analytical and simulation results are strongly correlated, which demonstrates the appropriateness of our analytical model.