The Main DepthStencil buffer is the one used when rendering to the Main window and thus will have the same resolution as the Window. The shadowing process however doesn't have to use the same resolution, viewport and so on. That is why they create a second DepthStencil buffer that has the same resolution as the viewport that is used for rendering the Shadowmap.
Also, as the quote states, it might not be desirable to use the same Format, Multisampling settings and so on that you use to render your scene, to create the shadowmap.
Edit:
You use the stencil buffer because you want to store the depth of the nearest occluding triangle.
A DepthStencil buffer doesnt "support shadowmapping" its just a technique that happens to need a depthbuffer, preferably with certain attributes (like right size, no MSAA and so on)