Pruning the computation of distributed shortest paths in power-law networks