You really need a large volume of repeated results by different groups doing the experiment/research so you get the proper regression to the mean. Individual papers are more important at saying "here is something interesting that others should also check out".