I first started with projection shadow mapping, but it has many problems. When the light direction and camera direction are close to matching, there's no way to skew the perspective to get everything into the view, and you get very very low shadow resolution. After playing with this for a long time, seeing the same problem in NVidia's demos, and lots of reading, I decided perspective shadow mapping is only good for academic papers, and cascade shadow mapping is much more robust, which is why everyone uses it.