I would say that the latter is non-optimal, i.e. generates some unnecessary intermediary values whereas the former is optimal as shown in the paper.