It seems to make sense that calculating an intersection point takes longer than traversing a ray, but how is the C_trav ratio figured out? In what cases would the ratio be closer to 1:1.5, and in what cases would it be 1:5?
In practice, one would likely benchmark rendering a number of scenes with a range of c_trav values and choose the one that work best. An acceleration structure where visiting a node was relatively expensive (e.g. a bsp-tree with an arbitrary aligned splitting plane vs a kd-tree with an axis-aligned splitting plane) would push the ratio up, while more complex shape intersection algorithms (e.g. an iterative intersection solution, like hw2) would push the ratio lower..