There is an even better way to look at this to prove it more convincingly. If you think a little harder about the test blurbusters did you realise we can use the minimum latency they saw. Why you ask? Because it was sufficient time for the mouse button to be clicked, led to light and then see the impact on the screen, it is a full and complete latency picture. The variance is all to do with timing, there is an 8ms (at 120hz which is the right one to use as 144 is being impacting by vsync waiting) difference in frames and if the mouse is clicked at the worst time it can add up to 8ms of additional time. Thus we see a variance. There is other variance there as well, likely due to the game CPU usage being somewhat uneven but in essence the minimum is as valid a number as the others and we don't care about the average, the average is hiding additional latency from the timing.
This is an okay idea, it does however have one weakness, namely that fact that it assumes that the measured minimums for vsync and G-SYNC, are at the global minimum, which isn't necessarily the case. But even then it should still get us closer to finding the actual latency.
So now the number that has to hide 7ms is 17ms. The GPU is taking 8ms because we are at 120hz (1000/120 = 8.3) and tests of this panel in other places tell us pixel switch time is 5ms, grand total so far being 13ms there is now only 4ms for the CPU, mouse sampling and most importantly there is meant to be an additional 7ms of screen latency in there. But there isn't.
That's not how an fps cap works, the GPU doesn't suddenly become slower and take longer to render a frame, instead the GPU (or more precisely the CPU, before it passes the next frame to the GPU) is made to wait before starting on a new frame until enough time has passed to stay under the fps cap.
Since we know that the titan they can run at least 300 fps (and probably significantly higher), it render a frame in an average of 3.33 ms. However since where looking at the minimum latency time, chances are that the render time is actually below average as well, so most likely there's only 2-3 ms of GPU render time.
The time spent by the CPU on the frame adds up to the total frame time and since we know the average frame time when uncapped is at max 3.33 ms, the CPU most likely doesn't spend more than 1 ms on average per frame (unless it can start on the next frame before the GPU has finished, also known as a render ahead queue, but I don't know if CS:GO uses that).
So that gives 2-3 ms for the GPU, 1 ms for the CPU and 5 ms for the panel. leaving up to 9 ms for latency (from any other source), and thus still enough room for 7 ms latency from a single frames worth of latency.