Yes. Given closed shapes, it works well in most cases.
One anomaly is not noted in the paper though. The second-depth (back-facing) polygon must not touch the third nearest surface, which should be in shadow. Near the intersection of the second and the third shape, onto the third surface falsely lit pixels occur, which are rather disturbing.
it is as perfect as you want it to be… with a nice resoluton (16bit or 24bit, for example), it is very very accurate… why? cause the MAJOR problem is finally solved, the front-face-flickering… the chance that you can see the faces wich are that near to the backfaces and that you then can see the flickering are very very very low compared to the normal shadow-buffer-problems… so we can say it is solved… that way you get very accurate and fast and simple shadow-mapping…