It might be simple attachment if a character is using skeletal animation, eg Intrusion 2. That art style isn’t used often because the direct limb tweeting is often overly visible. Often, most character frames are hand drawn or at least prerendered.

In these hand drawn styles, a character’s head could appear to enter Z depth as part of the drawing (imagine a 6 frame animation of a character spinning a sword like a top). When that happens WHILE they’re also wearing an attached hat, the hat must rotate and adjust for the depth as well - which means new drawings, even if you’re able to specify the positions of the character’s head during each frame of the animation.

We could be talking past each other with bad descriptions that need visuals, though.

With 3d you make the model and it's "naturally" 3d (obviously). If you want to make a 2d sprite have a different perspective, you need to animate (often times draw) it specifically. As they mentioned it before, it's mostly useful for animations and movement. It may not even be "reusability" as much as "lack of need to think about perspective" or "scalability".

Another point is that with a 3d engine under low-storage concerns (like say, the N64) you can do a lot of fuckery like having a total of ~10 textures and just apply various color tints (and maybe a blur here and there) to make it seem like there's more. While 2d engines do support this nowadays, it's still hard for artists to "fake" such a wide gamut of sprites, just by the nature of the medium. There's no model to apply a texture to, so you're limited to having a base sprite and recoloring it.

You could do a modular approach in 2d. For example, a character is built of the body (arms+face), hair, pants, shirt and shoes and change them individually. Same for houses with roofs, doors, windows and walls, etc.

However, as already said, you're limited by perspective a lot. Each new perspective requires almost double the sprites.