In this paper, we first present an exploratory analysis of the aspects of multiple-version software diversity using 36,123, programs written to the same specification. We do so within the framework of the theories of Eckhardt and Lee and Littlewood and Miller. We analyse programming faults made, explore failure regions and difficulty functions, show how effective 1-out-of-2 diversity is and how language diversity increases this effectiveness. The second part of the paper generalizes the findings about 1-out-of-2 diversity, and its special case language diversity by performing statistical analyses of 89,402 programs written to 60 specifications. Most observations in the exploratory analysis are confirmed; however, although the benefit of language diversity can be observed, its effectiveness appears to be low.