OK, lots of questions....
I would like to see a statistical calculation that shows why that is not true.
Good question. Let me answer it this way. It's kind of a backwards example, but maybe it will do: All the random variables that add up to group size add by the square root of the sum of the squares. So assume that you are offhand shooting a perfect rifle, and getting groups where the standard deviation of shots from the center of the group is 2". The variation is 100% from your marksmanship.
About 95% of your shots will fall within plus and minus 2 standard deviations, so plus 2 standard deviations is 4", and minus 2 standard deviations is another 4", so that's an 8" circle that will contain 95% of your shots. So, for practical purposes, that's our 8" groups.
Now assume your technique remains constant, and you switch to that rifle that shoots 2" groups off a bench. The random error due to the rifle has a standard deviation of 1/2". That is, plus 2 standard deviations is 1", and minus 2 standard deviations is another 1", for a total of 2" where 95% of shots will land.
How much does that degrade our 8" groups from a perfect rifle?
square root (2^2 + .5^2) = 2.059". 2.059" x 4 = 8.23"
Using our plus and minus 2 standard deviation rule, that says that our groups will grow from 8" to 8.23" by switching to the inferior rifle.
Most people have a hard time getting their heads around the concept, but it works that way because the probability of all the contributing variables happening in the same direction at the same time is small. Sometimes the errors are in opposite directions, and partially cancel, and sometimes they are in the same direction and partially add.
As other posters have pointed out, in this made up case, by far the largest source of variation is the shooter. Working on charge weight, neck tension, etc. isn't going to improve the results. It's practice, practice, practice. I deliberately chose that example because it illustrates the futility of working on the smaller sources of variation. An example for a bench rest situation would be a different situation, because that would eliminate the single largest source of variation.
He’s ONLY correct if and only if the variability exhibits a normal distribution
The square root of the sum of the squares works for all real world distributions, not just the normal distribution.
In reality powder charges don't always increase velocity in a linear fashion.
Actually, over the range of normal loads, MV is very linear with respect to charge weight. I find this puzzling, because intuitively I would expect kinetic energy to be linear with load. But my intuition is wrong on that one. I had a lovely data set from Ken Oehler, and when I ran the math, my mouth dropped open. The linear correlation was tighter than any relationship I had found in the reloading world.
27fps is an extra 8” of drop at 1000 yards for a 308win.
Not relevant. What was presented was a change from a 25 FPS standard deviation to a 26.9 FPS standard deviation, not a 26.9 FPS shift in the mean. That's a different proposition entirely. If you follow through the example I worked out for Bill M, you can find out just how egg shaped your groups become for a given rifle, for given MV standard deviation. It's not much, even for very fine rifles.