Ah I understand it better. So you're wondering why people would have mention "marking" without implementing XEP-0394? I actually agree with you, I think that this was suggested (not by me initially) as you say because people don't want to implement the whole marking thing (with bold, italic, etc) just to get mention. However, people can do partial implementation if they want.
A partial implementation still lends "legitimacy" to the XEP and increases the odds it ends up at least partially entrenched in implementations, though. And this would hardly be the first time we've done this. Should fallbacks be part of 0394 too? Nevertheless when I look at the actual example given, maybe 0394 could be split in such a way that we get a XEP defining the <span/> element without the <markup/> parent on any defined children. Then it would of course make sense for mentions and fallbacks to use that. Instead of moving everything that uses a span concept into the same one XEP.