Curious how much of the savings are because of actual core language processor efficiency vs the cost of garbage collection on systems which do a lot of memory allocation.
Too bad the article stays shallow on details of where time was actually being spent, you'd think they would have measured and reported on that.