A new implementation of a force decomposition method for parallel molecular dynamics simulations is presented. It is based on a geometrical decomposition of the influence matrix where sections are dynamically reorganized during the simulation in order to maintain a good load balance. Furthermore space filling curves are used to sort particles in space, which makes memory access more efficient and furthermore reduces communication between processors due to better locality. Benchmark runs are presented which shows in improvement in scalability due to well balanced work on the processors.