Emission tomography is among the most important imaging modalities for tumor diagnosis and visualization of metabolism. Image data are reconstructed from projection data of a radioactive tracer distribution within the human body. A novel fully 3D iterative reconstruction technique was developed to exploit full SD information obtained from projection data. Using a stochastic beam model both scatter and scanner geometry are implemented, providing improved accuracy in contrast and resolution compared to standard reconstruction algorithms. The algorithm needs high computing resources and was implemented as a service oriented system on high performance computing platforms. The codes running on SMP clusters may be exposed as Jini services, as Web Services or as OGSA services accessible over the Internet by means of a browser-based GUI. The system provides basic support for dynamic service management, service selection and monitoring based on Jini, Web Services and OGSA.