Mesh attachments use a little more resources all around than region attachments: file size, memory, CPU to render. The difference is small, but generally you'd not make a region into a mesh unnecessarily.
It's fine to bring your images into a skeleton like this, but why would you want to use images that are too large and then scale them down? Those images will take up more space on your texture atlas. If they are rendered scaled down at runtime, there is no gain in quality. Actually quality usually suffers somewhat when images are scaled up or down.
There are cases where you'd want to use higher resolution images, such as when you plan to render the skeleton larger on screen, like when zooming in for dialog/whatever. In that case you could still rig in the editor at the normal size, then use a texture atlas with larger (or smaller) images at runtime. Or you could use meshes, as mentioned.
