Really, this is a question of a reduced attack surface. And this may seem somewhat counter intuitive at first. Surely distributing your data across multiple nodes *increases* the attack surface?
Yes, on its own.
But the biggest problem with centralization isn't that a centralized system is easier to breach. It is that the expected payoff for a breach is much, much higher than when attacking a smaller system. The motivation to attack centralized systems thus increases.
Appropriate at-rest encryption - that is, end-to-end encryption - then ensures that breaching a smaller node does not automatically result in a leak of all data stored there. Only the subset of data for which there exist also compromised keys can be retrieved.
Add appropriate obfuscation techniques, such as avoiding leaks of metadata or mixing junk data in with the real, and the payoff shrinks even further.
This, roughly speaking, is the point of view from which my #interpeerproject operates.
A private instance for the Finkhäuser family.