You are missing the point - sure a particular enzyme's function is resilent to large levels of substitution because:

1. The number of residues actively involved in catalysis might be small and 2. Most other residues can be safely replaced with something else either similar if part of the structure or anything if the side chain is pointing out on the surface.

However, the point the article is making is that for different functions the same basic folds seem to be used again and again.

Is that because the stable protein fold structural space is actually small ( due to the limited secondard structure patterns used etc ), or is that because evolution hasn't had time to to search the enormous available structural space?

ie is it a sampling problem or an instrinic property of protein space.

The fact that some of the ML approaches mentioned can now design completely novel folds suggests it is at least partially a sampling problem.

This to me isn't surprising - the idea that evolution is somehow complete and all possible solutions have already been explored seems to me to be unlikely - a lot of evolution happens via gene duplication and then gradual functional drift - which would favour reuse of existing folds over the generation of completely new ones.