Quoting "Gastón C. Hillar":

This performance waterfall will help us understand how we can take full advantage of modern multiprocessing. The topmost part of the waterfall represents the best performance.
Hence, we lose speed as we go down each step. It is not a linear relationship, and the hardware infrastructure in which the application runs will determine the exact performance loss with each step represented in the above figure. However, the cascade is the same for every case, neither dependent on the kind of application being developed nor the hardware being used.

Details in the book: C# 2008 and 2005 Threaded Programming Beginner's Guide.

Cool!