I agree, I'd even go as far to say that not publishing non replicable results is better for research than publishing it. These papers are noise that is pointed to for marketing purposes. A higher signal-to-noise ratio will lead to faster research, because people don't have to sift through 10 papers to find one good one.