Systems, methods and computer program products for improving overall end-to-end runtime latency of flow graphs of message transformations which are placed onto an overlay network of broker machines by aggressively replicating stateless transformations. Exemplary embodiments include a system having a process including defining a message transformation graph, receiving information about measured and estimated properties of a message flow, receiving information about physical brokers and links in the overlay network onto which the message transformation graph is deployed, labeling each of a plurality of stateless transformations associated with the flow graph as replicable, heuristically determining a number of replicas and a corresponding load partitioning ratios among the number of replicas for each stateless transformation, converting the message transformation graph into an enhanced flow graph, running a placement algorithm with the enhanced flow graph and consolidating each of the plurality of virtual replicas that are assigned to a common message broker.