According to this: https://www.youtube.com/watch?v=BTURkjYJ_uk Firefox's servo engine can compose CSS elements/Display List together at 500 frames / second. Maybe next version of Windows / Linux desktop should use FF's servo engine?
You could potentially reduce this delay as well by having the application and the compositor in communication. Since rendering is going to be synced to vblank if you can get the application to not try to sync as well and instead just notify the compositor when it is done drawing a frame you could potentially get the application drawing and the compositor drawing in the same vblank interval. This is what Wayland and DRI3 Present let you do in the Linux world, I assume Windows has something similar but you'd need to opt-in to it so I bet nothing uses it.