I don't think this aligns with the Chinchilla scaling law. There is indeed a point at which you can oversaturate a model with data, as well as such thing as not giving it enough.