Well, it's also a question of how far abstracted your high-level API is. Of course you cannot provide a tesselation shader alternative, if none exists. But seeing that this is usually an optimization which will be enabled or disabled even for those platforms that would support them, there's no reason why it would be just bad to have some lacking support here or there. And if things are slower of less detailed due to old hardware that's to be expected by the users anyway.
@fmx: Yes, for me a basic implementation with, say, support of 3/4 of the features shown in the Irrlicht SDK examples would be enough to put it into the core engine. Before that I'd probably host it as another development branch just as the ogl-es drivers (which are to be included into the core in the near future, though, as they are rather feature complete).