Section 8 "Component Recursion" talks about a mechanism that can be used as addition to an individualization-refinement algorithm improving their performance. It is said that this mechanism is implemented in the newest version of bliss. I guess this is related to the "NON-DISJOINT UNION OF TRIPARTITE GRAPHS" benchmark graphs, where bliss, nauty, conauto, sausy, and Traces show interesting differences in behaviour between automorphism and canonical labelling (and between the different solvers).

pallini.di.uniroma1.it/Miscellanea.html

Despite the interesting fact that this component recursion can indeed be implemented efficiently, the chapter just shows that this is irrelevant for the given example graphs, since they don't have non-trivial components.