What you're observing is that most research results don't have much impact. The ones that get replicated are definitionally the ones that matter the most, since they lead to controlled trials. In principle this is a pretty reasonable way to allocate limited resources.