Tessellation shader question

When writing a control shader I can specify “patch out” on some stuff and “patch in” in the evaluation shader.
Since all output vertices (non “patch out”) from the control shader are duplicated for each new vertex from the tessellator, why wouldn’t I use “patch out” which doesn’t get duplicated, and run the control shader only once.
Seems like a faster (and memory economic) alternative.
Yeah, if the patch gets too big then it’ll be a long control shader, so let’s rule this option out.
Is the reason not to do it has something to do with transform feedback, or when there’s no evaluation shader?