I didn't write the article, and I'm not much of an expert on DS graphics (it's too surreal for me). But you can copy 3D content to 2D images and then display it using the 2D engine, I believe, and then use 3D on the other screen through the direct method. Pretty much everything on the DS is a bizarre special case, and Nintendo's in-house staff is very good at exploiting every last trick in the book.
The article mentions that the Main 2D Engine had a mode that could write to a framebuffer that was the combined area of both screens, and could pull from the 3D engine into that full framebuffer layer.
Also, you'll notice in the New Super Mario Bros example that many of the sprites were pre-rendered 3D and the 2D engine was just pulling bitmaps at that point.
So a bunch of options with all sorts of interesting trade-offs and it would be interesting to see breakdowns of specific games and how they used the trade-offs.
