A switch architecture and method provides scaling through multi-level load balancing of flows across data and application processing planes. An input/output module receives a communication session flow (forward) from a client device and selects one of a plurality of data processors to process the flow. The selected data processor determines the level of processing needed for the forward flow and selects an application processor from a plurality of such application processors. The application processor generates a session structure identifying actions to be performed on the forward flow and transfers the session structure to the selected data processor to perform the actions on the forward flow. The application processor also predictively generates and offloads a session structure for the associated reverse flow. If the reverse session structure is offloaded to a different data processor, either the forward or reverse flow redirects packets, or is redirected, to the data processor hosting the other flow. Typically, the redirection is accomplished by offloading a redirection structure as the offloaded session structure.