Complementing what Diogo said, we could probably optimize it further just by doing some extra spatial partitioning. Or maybe doing some compute shaders (if we do a standalone build). I will eventually test how many tiny meeples we can run with some extra work put into this.