Smaller communities are easier to moderate. Reduced interaction with other communities means that any one person can only have a limited effect before a remote moderator intervenes.
What happens when a toxic community starts causing problems in the wider network? Quite simply, they get cut off from the network. Every individual user can block an entire instance. Each server can also block another instance. There is also precedent for ALL servers to block one particularly bad instance.
You probably can't prevent toxic communities from forming. Since it's an open protocol, you also can't prevent certain political parties from setting up their "free speech" instance and doing what they do. But you can cut them off from everyone else just trying to have a good time.
The community structure should tend to gather like-minded people together. That would mean some instances becoming intensely toxic. Which is what we have seen happen. On the other hand the isolated structure makes it easy to just ban that entire community with one button.