In a recent paper the first author developed a dynamic-programming-based heuristic for the two-dimensional pallet loading problem. This note presents improvements to that procedure which have resulted in reductions in CPU run times of up to 1/20, as well as drastic lowering of memory requirements. A fundamental improvement in the bounding function is developed, and another improvement taken from observations of real-world pallet loaders and furniture movers is presented. Other implications of the improvements are also discussed, and results are compared.