The GeforceNow client has its own delay in processing your input, just like the game does. So you have to wait for your input to be captured by the client, processed, sent to their servers, where it has to be processed again, then given to the game process and then used there. And then the visual output has to be sent back, where again it has to be processed and displayed, with all the latency of the local client and display device.

Like, I work in video games for a large publisher - we have done a lot of testing on this technology, both on GeForce Now and our own internal solutions. Without getting into details I probably shouldn't talk about - the lowest latency you can see when streaming is around 150-200ms, and that's in absolutely perfect conditions where you have a data centre super close to you. Unfortunately, 150ms is visible to your average player, and in our testing the enjoyment of the game is directly related to what kind of game it is - strategy games, action games, even driving games? Mostly fine, not really noticable. First person shooters? Extremely noticable, in our testing there is a noticable drop of player performance in online PvP when streaming. We have been experimenting with improving it(by giving the player just a touch of auto-aim assist when they are streaming), and we were able to bring the numbers back up to where they were for non-streaming players. But then you get into a debate about whether that is fair or not.

Either way, like OP said - you can't beat physics.