Experimental studies of social dilemmas have shown that while the existence of a sanctioning institution improves cooperation within groups, it also has a detrimental impact on group earnings in the short run. Could the introduction of pre-play threats to punish have enough of a beneficial impact on cooperation, while not incurring the cost associated with actual punishment, so that they increase overall welfare? We report an experiment in which players can issue non-binding threats to punish others based on their contribution levels to a public good. After observing others’ actual contributions, they choose their actual punishment level. We find that threats increase the level of contributions significantly. Efficiency is improved, but only in the long run. However, the possibility of sanctioning differences between threatened and actual punishment leads to lower threats, cooperation and welfare, restoring them to levels equal to or below the levels attained in the absence of threats.