That's not true. It would be like saying that some human reading tons of books will only have “average” writing capabilities.
Learning is an additive process. Specially with complex neural networks and the training algorithms used. When you learn something you don't necessarily need to unlearn something else to make room for it.
It's like LLMs having some kind of intelligence cap. But given that cap, they max it out and perform way better than the equivalent human.
That being said, most vibe coding quality sucks. But not because of the reasons you mention.