I think self-voting is not that problematic in linear system because self-voters cannot reward more than their stakes. But in superlinear system, they sometimes, especially combined with collusive voting and follow-up voting bots, have more rewards than their stakes.