The latency/jitter of 3d graphics pipeline is pretty large (and even larger for a game), such "single point sampling" isn't a quite reasonable idea.
The latency from user input to actual display can take several frames and is fluctuating. The way the game works (if it doesn't "lie") depends on measuring a uncalibrated "latency", which could be random on higher precision, and more likely worse: due to systematic errors, a distorted distribution that is much likely to result in some latencies than others, on a specific system.