In the 1995 Precision Shooting Reloading Manual, the late Dan Hackett, a benchrest shooter, described a 220 Swift he had that he couldn't get to shoot 5 shots into better than about 3/8", with most groups being more like 1/2" at 100 yards. Most of us would be pretty pleased with a gun that shot this well, but this is a benchrest shooter we are talking about, so he wanted bugholes. He'd always seated bullets for this gun 0.020" off the lands because it was common knowledge among benchrest shooters at the time that this was the best jump. (Note that it had been common knowledge among benchrest shooters a generation earlier that contact with the lands was best.) Well, one day, in switching his load to a bullet that was 0.015" shorter than the first one he loaded, he accidentally adjusted his micrometer seating die in the wrong direction, turning 0.015" deeper and setting these bullets a whopping 0.050" off the lands. He had 20 rounds loaded before he noticed his mistake. He considered pulling them, but decided to shrug it off and just shoot them in practice. To his amazement, they gave him two 0.25" groups and two true bugholes down in the 1's.
So, what depth is best? The one that works best in your gun. Berger has found that in some guns the jump needs to be out as far as 0.150" to get the best our of the VLD bullets. They have a
description of how to determine that best depth on their site.
How much better is best? Well, if you shoot at well as Dan Hackett did, then you can probably pick up pretty easily on even a small improvement. If you don't, you have a statistical problem. A typical target group's bullet hole distribution may be modeled by a pair of bell curves at right angles to one another, the random locations within which provide coordinates of the hole locations. Each curve has a standard deviation around the group center, and if the group is round, the SD's are equal. The error introduced by improperly tuned seating depth has it's own standard deviations that have to be added to the first two to get a resulting effect. If we treat Hackett's before group average as 0.5 inches and has after group average as 0.2", then seating to a better depth reduced a contributing standard deviation by 0.458". It's bigger than you'd expect because standard deviations interact such that they produce deviations equal to the square root of the sum of their squares. So, suppose your gun shoots 1" groups and your remove that same 0.458" SD from them, what will your new group size be?
√(1²-0.458²) = 0.889" or barely over 10% reduction. Can you even see a 10% reduction reliably? If you miss the tuning spot by 0.020" and that number increases to a 5% reduction, will you be able to tell?
That's the issue. You may or may not be able to tell. If depends how much the error contributes and how well the gun shoots with the error in force.
Incidentally, all fine tuning tricks are like that. Deburr flash holes or not? Outside turn necks or don't? All these kinds of improvements can be apparent in one gun and not in another depending how large other error sources are.