Alignment and distribution of array data should be managed by optimizing compilers for parallel computers, but current approaches to the distribution problem formulate it as an NP-complete graph optimization problem. The graphs arising in applications are large and difficult to optimize. In this paper, we improve some earlier results on methods that use graph contraction to reduce the size of a distribution problem. We report on an experiment using seven example programs that show these contraction operations to be effective in practice; we obtain from 70 to 99 percent reductions in problem size, the larger number being more typical, without loss of solution quality.