The flaw isn't that it is not realistic enough, but that it goes too much against an average human's preconception of how a basic traffic system should work.
It is absolutely not unreasonable to expect a 1993 game to not choose a random route at a junction; it should use a basic pathfinding algorithm (which were well-known, even in 1993) to choose the closest one.
Of course it is limited by performance but optimization is hardly a good counter-point to intuition in this kind of game.
What if the flawed routing simply exaggerates flaws with the way we think it should be designed?
The number one bottleneck in real traffic is intersections. Some people think optimal city design should mix zoning types at a finer level.
To your point though, we can't really demonstrate what's better when the agents are acting stupidly. I'd like to be able to learn what would work in the real world too.
Yeah, also it is not clear to me why is that so bad, so what that you have a circle line, wouldn't only a microscopic fraction of paths get stuck on it after a large number of crossroad steps ?
This game was meant to run on a 25mhz 386 and 4mb of RAM. Complex pathfinding and over-complicated sub-simulations would have drastically limited how large of a city you could build before your system just stopped working.
It is absolutely not unreasonable to expect a 1993 game to not choose a random route at a junction; it should use a basic pathfinding algorithm (which were well-known, even in 1993) to choose the closest one.
Of course it is limited by performance but optimization is hardly a good counter-point to intuition in this kind of game.