I understand that you ask me this two questions. It may had two meanings
1 performance of NAIVE should NOT better than INSTANCING
2 showcase “webgl_instancing_performance” work well on your computer ?
It’s not a particularly scientific test right now, there might be better or more realistic ways to set this up. I wonder if a scene that contains a mix of different objects (rather than 10,000 of the same thing) would have more GL state changes and worse performance in the naive case. And of course the hardware implementations do vary a bit.