Years ago there was a fascinating study/experiment. They would put the subject in charge of maintaining the temperature of a room within a specific range, and the only action the person could do would be to press a button to cause an increase or decrease of the temperature; they could press as often as they wanted, but the change would only occur 5 minutes later. What happened? Basically everyone failed, consistently overcompensating one way or the other when they would see the temperature go up or down based on decisions made 5 minutes earlier. The more it went, the more people used the button, and they only made things worse.
Same thing would happen with this kind of solution. Decisions take time to be implemented, and by the time the side effects are known, more decisions are already in the pipeline.
For instance when you start sending traffic in local streets, local people react; they leave earlier or later, causing changes in the rush hour patterns. Or they fight back, getting the local authorities to convert one-ways to two-ways, or to reverse traffic flow. Thru traffic is slowed down, this causes new bottlenecks, sending more cars on other streets. Rinse and repeat. And while this happens does the computer model take the big picture into account? No, it keeps rerouting people in a progressively more frantic way until the whole area is a gridlock.