You shouldn't notice lag. On modern Xorg the only round-trip is context switches between server and compositor, because the only thing what is shared is texture dma-bufs (there is inefficiency in mesa code for GLX_EXT_texture_from_pixmap extension, but it is other story). And if dma-bufs is working (Xorg needs to test and pull one MR) you have buffer direct scanouts as in wayland.