So lets evaluate the numbers, by taking in account the theory you were trying to evaluate.

Your attack and ninjutsu in the tests was: 14750/13557.

The enemy defense and resistance was: 16246/15523.

Anbu special tactics grants +20% atk. Earth main passive grants +40% def +20%res.

So the real stats of the fight were:

Your ATK/NIN: 17700/13557.

Enemy Def/Res: 22744/18627.

So the difference between them were: def-atk = 5044. res-nin = 5070.

So by adding in the p3netrations in the first case you reduced them to: 400 and 936 in the second case to 1471 and 1760.

The difference between the two tests were a mere 1071 less defense and 824 resistance.

Lets add in the fact you are using midnight blade, whose attacks are all lightning elemental based (so comes in effect the elemental resistance of the enemy too to mitigate the effect) and that almost for sure the enemy had not less than 2k damage reduction and what do you get?

That the difference in real damage landed could easily be within the rng range of the damage.

(as a pure example, by taking in account a full tai effect with those stats, since def effect very likely go with the square root, just to tell you: 17700/(22344^0.5) = 118.4 and 17700/(21273^0.5) = 121.3. Lets say the total multiplier in your case is 7, the difference in the average damage with those stats was 829 vs 849 ...)

I just told you last time that midnight blade is not the best way to test those things and even if you thought the stats you were using were fine for the test they actually are not.

Next time you want to test tai pen use first standard attack of water main or earth main's.

Next time you want to test nin pen use fire main or wind main 10 combo attack and be sure the enemy doesn't own fire or wind elemental resistance.

That said, the lower the damage you land the lower the effect is.

P. S. My guess is that the penets are always there, not only for when your atk and nin is lower than def and res, just that on very stacked defensive lineups they do not matter much while they work as intended when you face the classic overstacked move 1 that fully forgot to stack move 2/3/4, since the lower the def and res the higher the impact of penets is.

@ICEx: to test my assumption tell to yugaotamura or to crofghter (don't know who you tested with) to go naked with his move 2/3/4 and look at the damage you land on his move 2/3/4 with and without the inner gates runes. Just use your common edo hirzuen nuking meta. Imho the damage, in average, will be way wider on those moves if you use the pen runes while will be just slightly higher on his move 1. Obviously this is an extreme test and not really useful for real battles but is the best way to show if they really do something or not, since you should be able to nuke them with way higher damages.