The partitioning strategy described in this paper provides a mechanism for balancing the load of three-dimensional iteration spaces. Comparing to cyclic schemes of partitioning [5], the advantage of the presented scheme is that it is based on forming groups of consecutive iterations, an approach which is likely to reduce false sharing effects on certain machines [8], [9]. Furthermore, the presented scheme partitions the iterations of the outermost loop into equal (or almost equal) parts (as opposed to approaches analogous to balanced chunk scheduling [7]), which may be preferable for exploiting data parallelism at the program level.