Yeah, definitely aware that it's slower - I was severely bounded my memory when I came up with this, as I was trying to get my training data and all the variables in my model into 8GB of RAM for a machine learning project. Without the trick, swap was killing the speed; with the trick everything sailed along :)