Consider calibrated 10% ballistic gelatin as a homogenized human. Like all simulants, it MUST be homogenous in order to be both useful and valid. Originally, it was only meant to simulate pig leg muscle in order to placate the animal rights activists in an era when we shot anesthetized live pigs, but it turns out to work well as a human tissue simulant.
Bullet penetration in actual humans is quite variable, sometimes to a striking degree, often without apparent reason. The newer designs are far more consistent than older designs, but there is still enough variability that arguing over whether HST is better than Gold Dot, etc. is like 2nd lieutenants arguing seniority.
I have had cases of multiple GSWs where FMJ under-penetrated JHP, and others where two identical bullets out of the same gun, passing on similar trajectories through the same body cavity of the same person, and striking essentially the same tissues throughout, have had wildly different penetrations (on the order of twice as much), despite looking pretty similar when I dug them out.
That caveat aside, gelatin IS very useful, for the primary reason already given above: it allows you to compare the likely performance of an unknown load with one whose performance is well-documented.
Where it is NOT terribly useful is in taking a gelatin-derived penetration depth and presuming equal penetration in a human body. We see those sorts of statements all the time justifying poorly-performing loads: "yes, the Tactical Buzzsaw Ninja load only got 6 inches of penetration in (uncalibrated) gelatin, but the heart is only a few inches deep in the chest; that's all the penetration you really need.", etc.
It's also not terribly useful if you only fire one shot of each load, but that's something for another argument.