The workload distribution approaches used by software distributed shared memory (DSM) clusters always distribute the working threads of applications according to the computational power of processors. However, in addition to computational cost, the cost of memory accesses is an important factor for determining program performance. Neglecting this cost will result in making wrong decisions in workload distribution and then degrading program performance. To address this problem, we propose a new approach with simultaneously considering the memory capability and the computational power of processors for workload distribution on software DSM clusters in this paper. We have implemented the proposed approach on a test bed. Our experimental results show that the proposed approach can provide more performance improvement for the applications compared to the others with considering only computational powers or memory capabilities.