This would be more complete if they had also done some benchmarking of the two methods. It might not produce any speedup at all given the complexities of recent processors.