That number was calculated by taking a high-speed camera and watching how long it took the line to catch up with the pencil. They cheat on the metric by juicing up the prediction. You can see this by flicking the pencil and then picking it up -- the line will go farther than you intended. And if you change directions, the line will take more than 9ms to catch up with you.
Apple are cheating the 9ms lag measurement the way speculative execution and aggressive cache population are cheating performance benchmarks.
In general, this point makes me think of computational photography. If you have extra computing power, you can do things that are synthetic, but nevertheless real enough to deliver satisfaction, in real time.
