It's because most scientists and the organizations funding them are not interested in advancing our knowledge of nature so much as looking like they're advancing our knowledge of nature. If we want to be less wrong, we need more replication. Scientists don't want to do replication because it advances our knowledge of nature without looking like it is. So instead they keep trying to find fainter and more obscure effects, so faint that they exceed their own abilities to recognize statistical significance. It fools reviewers, journal editors and employers though, so they keep it up.