>should drastically reduce our confidence in any generalization of even the seemingly more relevant result.

Oh, absolutely! I certainly didn't mean that you should go trusting the scientist to vet his own research; most studies say exactly what the researcher wants them to say. But I think the results are strong enough to prompt (probably stronger) testing in other labs.

If the research was done properly, then instability of results under minor variations generally implies that the variations weren't actually minor.