What I don't get is that people claim that the incentives are skewed because highly cited paper get you the top jobs. However, assume that a significant subset of the citation are citing because they require the fraudulent result, then this will increase the chance that it would be eventually exposed... and quickly.

That is assume that person publishes results that: "factor X seems to lead to outcome Y". Many other scientists will then start trying to establish the low-hanging fruit result: "something that looks like factor X seems to lead something that looks like outcome Y". In other words they will be performing a sort of replication but in a novel way. If the result is fraudulent, then none of these results will materialise. In other words I don't get how a paper can be fraudulent AND highly-cited without escaping scrutiny, unless we are talking of a fraud mafia.

Here I am using the field of pure mathematics as a mental model. Assume a person publishes a mathematical result with a flawed proof that escapes scrutiny. If this result is used by sufficient number of mathematicians (especially the lemmas used to prove the theorem) then fairly quickly it will end up generating self contradictory results.