Let's see a cite to that data, because it seems ridiculous. Obviously if you make a road less congested people may start preferring it to other, more congested roads, but that just means the reduction in congestion is distributed rather than concentrated, and how is that a bad thing?
The entire premise of this article is questionable. The theory is that people will drive more carefully when the roads are more dangerous. From this the argument is that we should make the roads more dangerous in order to make them safer. This is obviously completely counterintuitive so they point to some evidence that it could be true (which itself seems to imply that its veracity lies near the statistical margin of error). The real problem is that it's a false dichotomy. Making the roads more dangerous is not the only way to make people drive more carefully. And alternative ways of encouraging people to drive more carefully are obviously preferable to a method that, at a given speed and level of care, is more dangerous according to the laws of physics.
More than that, this “make the roads dangerous to make them safe” is completely backwards when we're on the verge of commercially available self-driving cars. A self-driving car is obviously not bound by human psychology, so for all of those vehicles, making the roads more dangerous will only make the roads more dangerous.