IMO, here lies the problem. I prefer the stability of content over the stability of definition. If I ever look at an older paper, and read that members of the clade A+B are mainly long-necked, I would need to check what taxa were at that point though to belong to A+B, as it might be that nowadays, because of the change in content, most taxa in A+B are known to be short-necked. Not a problem if you add or delete a species here or there, but if content changes much, a new name is better than a 'stable' name that can move back and forth through a tree. And if the grouping is found to be non-monophyletic, just forget the name used for it.