The technical term for this kind of markup is "stand-off markup" or "stand-off annotation".

There is wealth of research done on stand-off markup since the SGML times, with a renaissance in 2005-2010.

Historically, the main drawback (as highlighted by other comments) is that stand-off needs a way to refer to the content, i.e. a pointing mechanism. Pointers in a pointing mechanism can be either a) hard to write but easy to maintain or b) simple to write but prone to breaking after a change (and thus requiring support from authoring tools). Nobody has found a viable ergonomic balance yet.