Choreographies describe the interaction behavior of processes at design time: a choreography defines when messages have to be exchanged between the involved processes during their runtime. In the context of Web services and the de-facto workflow language BPEL, SOAP is used to encode the messages. When complex messages are exchanged between the processes, this can become costly and time consuming with respect to the overall execution time of a choreography. In this work, we suggest three different performance optimization techniques for workflow engines to reduce the number of message exchanges between the interacting processes and hence, to decrease the execution times and costs of the choreographies: intra-engine transport, service request caching, and inline execution. We describe how these techniques are implemented in a workflow engine. Performance measurements are carried out to determine the performance improvements that are achieved with each optimization technique. We further show that the optimizations also affect the energy consumption of the workflow engine.