This paper studies the streaming of packetized media from multiple servers to a receiver over a lossy network with retransmissions. Considering the packet interdependencies, decoding deadline requirements and diversified network/server conditions, we propose a receiver-driven media streaming framework that requests the most "important" packet for transmission at each request opportunity in order to minimize the end-to-end distortion while conforming to the rate constraints dictated by the flow, window and congestion control mechanisms. Simulation results demonstrate the benefit of exploiting server diversity with our proposed media streaming framework.