Is that extra latency an inherent limitation of how Wayland is designed, or could future optimizations improve Wayland so that its performance is on par with X?

Yes, it is an inherent limitation of Wayland because Wayland assumes a compositor and compositors run asynchronously to the applications that produce graphics (it could be possible to run them synchronously but then a single application could freeze the entire desktop - and even when it worked, you'd force all applications to sync with each other), meaning that what the compositor composes is based on the last complete frame it got (and while the compositor is composing its composition the application might as well be waiting, virtual thumbs twirling, with a new frame ready).

In comparison X11 doesn't assume a compositor - it can support composition, but doesn't require it. If there isn't a compositor, applications draw directly to the framebuffer and content appears as long as it is sent to the GPU (you may even send stuff faster than the monitor can keep up with - this is how you get tearing for example).

You can work around Wayland's limitation the same way you can work around it in Windows: get a monitor with faster refresh rates. As usual, better hardware is what fills the holes created by shoddy software (i got an expensive 165Hz monitor expecting to have a never-seen-before butter smooth experience on Windows - in reality what i got was more or less the experience i had back when i could disable the compositor in Windows 7).

I have a 144hz monitor and I can still notice the extra wayland latency; not so much in regular desktop programs, but in fast-twitch shooters like quake. And this is compared to X11 with a compositing WM. So I'm not sure if that's it.

I'm hardly a Wayland expert though, I was just commenting regarding my own anecdotal experience.

> Is that extra latency an inherent limitation of how Wayland is designed

No, most Wayland compositors can already skip composition and pass the window contents directly to the screen in case of displaying just a single fullscreen window. There is also work ongoing on utilizing hardware planes when possible, to skip composition and improve performance in other cases as well - going beyond what was ever possible on X. Grandparent comment is simply wrong.

Sorry but I have tried wayland multiple times (most recent ~12mo ago) and this has never been true in my experience. At least on my nvidia card. And the wayland people have been saying stuff like "oh we know better performance is possible, we just need to implement x, y, and z" for many years.