The Additive-Increase Multiplicative-Decrease (AIMD) mechanism employed by TCP hinges on its ability to meet specific bandwidth requirments since it has to respond to congestion signals promptly by decreasing its sending rate expoenentially. This, in return, imposes strict limitations on the achievable bandwidth required by a particular application, not just in terms of the number of congestion signals it receives, but also on the timing of when they do occur. To that end, this paper presents a new architecture, whereby a set of TCP connections (we refer to them as stunts connections) sacrifice their performance on behalf of another TCP conenction (we refer to it as the free connection) by observing this congestion signal and reacting to it instead of the free connection. With this architecure in place, applications can now request specific bandwidth requirments and have the free connection adjust which congestion signals it reacts to it and which to delegate to the stunt connections in order to meet those specific bandwidth requirments.