I have heard a suggestion that PHD eligibility should require a replication. This would not only help validate (or invalidate) studies but would also over time reinforce the importance of replication culturally.
I think we should make it more precise: in order to reach <stage>, you should have to replicate another's work that has not been replicated more than three times as of 6 months ago and produce either a record of a successful replication or a bug report detailing a failed replication.
As a best practice, a lab where few of the incoming grad students can produce a definitive (either successful or failed) replication should checked-on by the university to see if they've just got a terrible working environment.