In the realm of game theory, mixed-motive games represent a subset of games where the interests of players are not entirely aligned nor entirely opposed. This duality often leads the system to a state known as the collective action problem, when individuals systematically prioritize their own rewards as opposed to greater group rewards. This problem normally occurs in mixed-motive games in the real world because people are generally good at responding to individual incentives and, with the emergence of learning techniques such as reinforcement learning, so are becoming agents in MAS. In our previous work [8], we proposed a framework composed of several learning agents, whose actions were regulated by a regulator agent to prevent the collective action problem in mixed-motive MAS when the following two conditions are not guaranteed: a) most agents in the system, more often than not, act in favor of the group instead prioritizing their own rewards, and b) agents are allowed to inflict non-negligible harm to other agents in order to punish defective behavior. In this new work, we present two experiments in order to test the effects that two variables have on the system’s outcome; the frequency in which the regulator updates the system’s norm and the harshness of the punishment given to agents that violate such norms. We show that higher update frequencies and harsher punishments tend to yield better outcomes.