Facial animation based on joints is pritty hard to control-Imagine how complex should mounth joint chain be,how many bones to achive all the facial exprssions?

Blend shape will not satisfy your needs if you dont create enough blend shape targets to achive complex therefor realistic facial animation.

Problem that comes with blend shapes is that it is deforming linear in animation but you can add more blends in between to get more looking curve type deform(eyelid example).

Joints should be more accurate,but setting them up and skining is quite long and complex process.

Its not bad idea to explore joint based facial animation since I dont think that blends will hold on for a long time...