A system, apparatus, and method are directed to managing multiple back-end connections for pipelined HTTP communications. A traffic management device is configured to open back-end connections to multiple servers. The traffic management device distributes HTTP requests from a client device across multiple server connections. Instead of buffering the responses to ensure that each are returned in a same order as the requests, the traffic management device throttles a TCP receive window between all of the back-end server connections, but that back-end server connection associated with a first expected response. As each response is sent to the client device, the TCP window for the next back-end server connection is opened. This effectively offloads any significant buffering onto the back-end servers, enabling the traffic management device to return the responses in the same order as the requests.