I've gone in to this in depth elsewhere, but the difference is in the number of drive by bigots encountered, and how easy it is for those drive by bigots to interact.

So, in an open fediverse, with no defederation, and only user level filtering, drive by bigots will come along and hate on someone. That person will block the bigots instance (after being exposed to their hate), but even after that, the bigots will still be able to interact, and other vulnerable folk will still encounter the bigot dog pile. This is exactly how things work on twitter currently.

But when you can lock down posts to limit who can see them and defederate from instances that attract bigots, then less bigots will randomly see the content from vulnerable folk in their feeds in the first place, meaning less drive bys and less dogpiling. It doesn't stop targeted bigotry, and it doesn't stop dogpiling completely, but it makes it far more manageable.