BMTP offers rate controlled multicast with reliability, high throughput, and support for large numbers of receivers. A multicast sender needs feedback from receivers to recover from errors and to choose an appropriate send rate, but must avoid being overwhelmed as the number of receivers grows. BMTP does this by keeping the rate at which each receiver sends feedback inversely proportional to a running estimate of the number of receivers. BMTP bases its send rate on the minimum of the receive rates observed by the receivers, causing the sender to slow down in the face of packet loss or competing traffic, and to speed up when there is spare network capacity. BMTP’s NAK-based retransmission rarely sends any data more than twice, a substantial improvement over iterated unicast. Rabin’s Information Dispersal Algorithm can reduce this re-send rate as close as desired to the underlying loss rate of the network. Simulations with 1000 receivers substantiate these claims.