This paper describes an architecture and runtime system to implement distributed control and data processing applications in a thin-client manner, suitable for implementing a thin-client mobile robotics system. The system varies control fidelity and locality to adapt a control application to changes in Quality of Service availability and processing resources using a cost benefit model. An example application is presented in which the architecture is used to implement the distributed control of an inverted pendulum over a shared network. Performance results are compared with non-adaptive distributed control approaches.