Yeah, I agree that for the vast majority of things most people don't try. Or at least publicly demonstrate that they tried (key phrasing). But it is also silly to think that 3-5 people sitting at a desk reading a summary of work can validate said work. Really they can only invalidate or specify that it is indeterminate, but neither of these are validation. Which that's a key difference from the general public understanding of "peer review" (meaning journal publication).
But it might also be worth noting that often reproduction happens behind the scene. People point to big works like that which comes out of CERN, LIGO, or other massive projects and state that such works cannot be replicated. But actually those have high rates of replication, which is why there are hundreds of authors on the work.
For LK-99, people got to see a lot of what is typically done by grad students who never tell the public what they did (or even their community). That the communication between scientists is happening through preprints, email, twitter, and other methods that are not journal publications. Because science happens faster than the journal cycle. Most scientists are reading preprints, and letting the work dictate the signal of validity long before a journal can.
But what I was alluding to, which you might have picked up on, is that the reward system we have in place ("publish or perish", h-index, journals, etc) are misaligned as they do not reward this cornerstone of science -- replication -- (typically discourages is) unless there are credible claims of breakthroughs of the highest kind. Maybe we should rethink this system, and I hope that the timing of this along with the other discussions of academic fraud can help people to question the system and metrics that we use to evaluate, and ask if they are actually aligned with the original goals.