[PeterisP]

It's probably very different in different disciplines - in social studies like the ones in main article this is a big problem because, as you say, meta-studies will likely include these papers simply because they exist.

However, in more practical sciences if someone fakes data to show that their method A works better than baseline B, then other people building on that find out that method A doesn't really work well for them for a weird reason, shrug, and ignore the bad paper, so it doesn't get used and cited, while the correct assertions persist, get replicated, repeated and cited.