This work was supported in part by the Swiss National Science Foundation and by Cray Research, Inc. within the framework of the Cray Research/EPFL Parallel Application Technology Program. The computations were performed on the Cray T3D massively parallel computer at EPFL, Lausanne. The authors would like to acknowledge O. Sauter, S. Wuthrich and S. Merazzi for their help during the code development, T. Goodman for reading the manuscript and M.Q. Tran for very fruitful discussions.