The BBTI test has several examples that don't follow the pattern. The Fiocchi 125 grain HPXTP load actually shot slower in the 2" barrel with no gap than with one of either size. This was due to random velocity variation of the mean, which is pretty high with their sample size. However, I hacked their Excel file to run my own numbers on their data, and when you average all results for each barrel length or for all of them, the pattern is pretty clearly there. The first 0.001" of gap makes more difference than subsequent additional 0.001" increments. Longer barrels lose more to the gap than short ones because there is more time for gas bleed off.
All the 2" barrel results for .38 Special (I excluded the .357 and .38 LC data) averaged a loss of 1.5% from the first 0.001" of gap, and 4.5% from the 0.006 in gap, meaning about 0.6% velocity loss for each additional 0.001" of gap after the first one. Averaging all the data, which means an average barrel length of 10", the loss was 5% for the first 0.001" of gap and 0.7% for each additional 0.001".
The SAAMI standard test barrels all use 0.008" gaps. I don't know why BBTI didn't think to go with that number. I would have. But the pattern suggests the difference would be about another 1.2% loss with the 2" barrel.
The fact the most difference is made by the first 0.001" of gap explains why Hal would have trouble seeing the difference in his DW revolver using different shims to set the gap. I never noticed in on mine, either. But you'd have to be scratching the cylinder face up or else shoot a very large number of rounds to distinguish it from random variation with any helpful degree of confidence.