Distibuted Shared Memory (DSM) is a good solution to the scalability, complexity and high cost problems of large scale Shared Memory Multiprocessors, as well as the difficulty of the programming model problem of the message passing Distributed Memory Multiprocessors. We present a method for improving the performance of Distibuted Shared Memory Environments running on grid multiprocessors. The method is based on removing the inherent centralism imposed by X-Y routing that causes congestion in the centre of the grid. Simulation, as well as implementation results on a 1024 processor machine show an improvement of up to 24%.