It shouldn't be a problem. Lets say one 1920x1200 uncompressed color frame is about 10 MB. So, considering that x16 PCI-e link bandwidth is 8GB/s, it is possible to get GPU->CPU->GPU transfer speed ~400 fps (about 2.5 ms per frame).
BTW, Intel says that using MLAA algorithm it takes to CPU ~4.5 ms to process one frame at 1280x800 resolution and ~9.7 ms at 1920x1200 resolution (probably, including transfer time) and it doesn't depend on the scene complexity.
Bookmarks