Motivation: Structure based methods for drug design offer great potential for in-silico discovery of novel drugs but require accurate models of the target protein. Because many proteins, in particular transmembrane proteins, are difficult to characterize experimentally, methods of protein structure prediction are required to close the gap between sequence and structure information. Established methods for protein structure prediction work well only for targets of high homology to known proteins, while biophysics based simulation methods are restricted to small systems and require enormous computational resources. Results: Here we investigate the performance of a world-wide distributed computing network,