Performance similarities predict collective benefits in dyadic and triadic joint visual search