Ah, I was imprecise in my comment. I didn't mean that this implementation was doing the optimal ordering - I was just reminiscing about a similar project I worked on an why I abandoned it (I was unable to get the ordering done while keeping performance good enough with thousands of tiles