In the scientific field, the workflow technology is often employed to conduct computer simulations or computer supported experiments. The underlying IT infrastructure typically comprises resources distributed among different institutes and organizations all over the world. Traditionally, workflows are executed on a single machine while the invoked software is accessed remotely. This approach imposes many drawbacks which are outlined in this paper. To address these weaknesses we investigate the application of decentralized workflow enactment in the scientific domain. In this context, we explore the employment of process spaces, a middleware for the decentralized execution of workflows. Furthermore, we propose the combination of process spaces with the concept of data references to increase the overall performance of distributed simulations based on workflows. The considerations are discussed with the help of a scenario that calculates and visualizes the ink diffusion in water over a period of time.