An Xref is like a block, in that it acts as a single entity, but it is updated from the file when the drawing is opened. The geometry of the xref is not stored in the drawing it is referenced into, rather, it is read from the file and displayed in the current drawing and a placeholder, much like an insert, is used to reference the xref.
The xref is reloaded upon opening the drawing, thereby incorporating any changes made to the base xref file into the referenced file.
The xref can put into a drawing in a manner that will prevent it from being visible if the referencing drawing is inserted or xrefed into another drawing.
Layers within Xrefs can be manipulated independently of the layers within the host drawing, even if the layer name is exactly the same.
If I had to make a judgement, I would say that idea behind the xref is to make a way in which users can incorporate portions of their drawings into other drawings without the overhead of having to update them everytime a change is made to the base drawing. Also it alows multiple users to reference the same drawing and as a result multiple users have the exact geometry to work with.