What is very interesting about this is that it (probably) shows that the pure agent based ground up approach is completely impractical. I was very interested when the project was first announced as to how they would manage to do this, as complexity increases exponentially with size with an agent based system (which is probably part of the reason that the cities are so small). What this appears to show is that the very basic premise of the Glass Box engine is a fundamental design mistake, and that the conventional way of programming simulation games (as used by Simutrans, amongst others, of having a mix of high level statistical modelling and partly detailed ground-up simulation in key areas, such as, in Simutrans, goods and passengers) is superior, at least for present day hardware.
Incidentally, can anyone link to the video from 2012 to which Colonyan is referring?