In this paper, the authors analyze in theory how to increase the speed-up ratio of parallel test generation algorithm based on fault partitioning. The approach of backward fault partitioning of output fan-in cones (BFPOC) which combines the relevant fault recognition and shortest path sensitization, is presented. And BFPOC is compared via experiment with the approach of toward fault partitioning of input fan-out cones (TFPIC) proposed by Banejee and the general one, equal distance partitioning of fault sequence (EDPFS). The experimental results show that in large-scale parallel processing environment, BFPOC can reach higher speed-up ratio, obvious super to the other two approaches.