There are a few implementation issues depending on just how clever you want to be specifying a reference. Finding [id://nnnnnn] references would be easy. Finding <a href="xxxx">that node</a> references could be a bit mixed - it may be a bit harder to dig out the actual reference. Resolving [node name] references may be impossible - there are duplicate node names. Resolving "you know, that node by thingy, the wizard fellow, about sorting stuff" may well be beyond most humans let alone some code, but is still a reference, of sorts.

A second thing to think about is, when should nodes be parsed for references? Each time they are updated in the database? That gets a bit tricky avoiding bumping reference counts the wrong number of times. Something the vote fairy should do? She may get a bit grumpy at the unacustomed work (and you better watch out for the Vote Fairy Union too). Should all the old nodes be processed, or only new ones?

If a node is referenced multiple times from a single node does that count once, or once each time?

Having said all that, aside from the implementation issues, I quite like the idea. More nice warm fuzzies handed around and more opportunity for statistical analysis. Fun :)