Don't code for the future. Try focusing on what you need right now. What if you never get to a point where you need 10k+ instances on the screen because your games will be a lot simpler than that? You'd have spent [a lot of] time optimizing for a scenario that never came to pass.
And doing a stress test that has no stress in it (are you changing any render states? switching textures? drawing meshes with at least 100+ vertices?) also won't give you any (overly) useful information.

I'd advise against optimizing for the 1500 fps on a very simple case, and instead coding features that you need now, and optimize them when you notice slowdowns. It's hard to get yourself to that state of mind, but try. I wager it'll be good for you.