More of the Same: Persistent Representational Harms Under Increased Representation