I think the author is more interested in showing how to implement a certain problem using a GPU approach than to test a GPU approach vs. a CPU one.
As mentioned, the topic comes from an end-of-chapter problem set on prefix sums, at https://www.cs.cmu.edu/~guyb/papers/Ble93.pdf .
A prefix sum cam be implemented using a GPU, as demonstrated.
However, using a prefix sum may not be the best way to compute Fibonacci numbers. As demonstrated.
yea, i wrote this blogpost rather to show how to use scan in different ways than the canonical example of calculating prefix sum of a vector shown in introductions on gpu programming.