A character model is made up of "slots". The head slot, the chest slot, the legs slot and so on. When you equip a piece of gear, it replaced the body mesh in that slot. So a helmet model replaces the head, a cuirass replaces the chest, I think you follow. If you want a piece of gear to only partially cover the character, you need to create a new slot. But gear is easy to implement, since it conforms to the character's "body" and uses the same animations.

Now add a scarf. First, you need to create a new slot, so that equipping the scarf doesn't replace the head or chest. And then comes the question of animations. Are you going to have the scarf just lay flat against the character? That's the easiest approach, but it'll be completely static, look like ass and probably clip through at least some of your armors. You could use a cloth sim. If your scarf mesh has enough polygons, it'll look the best. But it's also computationally expensive, especially if you go with mesh-based collisions for maximum eye candy. And what types of objects can the scarf collide with? Just the character, or world objects as well? Every object the scarf collides with will create a whole new slew of physics calculations, all the time, dropping your performance in the gutter like a mob snitch. Or you could create a bespoke rig for the scarf. It'll look better than a static object and won't have a notable performance hit, but won't look as good as the cloth sim, especially since it won't collide properly with whatever else your character is wearing. And you'd need to create matching animations for literally every animation the character can possibly do. Every. Single. One. Your animators would want to murder you. And they will, when you come back to them a little later and say "Okay, real impressed with the scarf, now let's make 5 different ones. And I want capes."

TL;DR: It's not just another piece of gear.