I cannot explain the exact details but I remember during the first great Twitter exodus some people discovered a drawback in the ActivityPup protocol that seems to cause performance issues when very influencial users post on a small/under powered instance.

Because communicating all that stuff to many other instances is more costly than spreading it only to people on the same instance. So technically speaking large instances have a performance advantage and must just scale accordingly to the user number.

Everyone agreed that this need to change in oder to ensure a healthy federated ecosystem but I don't think it was be fixed by now.