This article was from 2009 too, and since then typed arrays have arrived which HUGELY sped up my messing-around-with-neural nets (without me actually having to write good code)!