Small effects multiplied by large populations, or small probabilities multiplied by large benefits, are very difficult to estimate and any estimates I can come up with have such large error bars that they are useless.

I should add that I think some of the ideas in the original post have this problem too. For instance, how would you calculate the magnitude of the effect by which saving infant lives encourages the demographic transition and thus reduces overpopulation?