Steiner systems are statistical designs that permit the comparison of all pairs of objects in a set in groups of three or more. Graph based evolutionary algorithms are a method of improving evolutionary algorithm performance by imposing a geography, in the form of a combinatorial graph, on the evolving population of solutions. The graph limits mate choice and information flow in the population. The choice of combinatorial graph that yields improved performance is highly problem dependent. This paper demonstrates that performance on the problem of locating difference sets that yield Steiner systems can be improved as much as 9-fold by using the graph based technique. For six different cases of the Steiner difference set problem the same graph yields the best result, suggesting that the problem is one for which the choice of best graph remains the same as the problem scales. Performance for the best graph versus that of a standard evolutionary algorithm is tracked beyond the six cases used and verifies that the improved performance scales. The results of these scaling experiments exhibit increasing many-fold improvement.