Now that we have explored both ways of generating facial animation in game characters, let's have a look at the issues, pros, and cons for both methods.
The big question concerns which version the game engine, or indeed the target platform, can handle. Chances are any one of them could use joints, since they will already be doing so to move the character around. You may even find that blend shapes are not eligible and need not be considered.
The way each type of animation works and looks can be drastically affected by your budget. Having too few joints, or too few blend shapes, will make the face look stiff. The freedom to include all the joints needed would allow you room to create an almost limitless number of facial poses and movements. With blend shapes, each new face would add to the overall size of the character in memory, so eventually you could reach your limit. Even so, this can be overcome through clever use of your shapes; you can mix existing ones to create new ones.
The initial setup of the joint-based face rig can be quite cumbersome and tricky, since you're having to use Set Driven Key to store each pose and area of movement. However, once they are complete, redoing them shouldn't be necessary, even if the character's topology is changed.
The blend shapes setup is simpler but does rely heavily on the main face's geometry being locked. Any alteration in the facial topology could result in your having to redo all the existing blend shapes.
With the joint setup, you can't get the facial expressions vertex-perfect, because each vertex's position is merely influenced by the joints. With blend shapes, on the other hand, you are physically sculpting each face, allowing you to get each pose perfect.
If you are lucky enough to be able to choose either method, it will come down to your own preference or style of working.