The success of screen-based faces are extremely dependent on two factors- flawless lighting conditions and properly sculpted facial surfaces (and of course proper maintenance on the projector bulbs/focus/etc). If either area is ignored or poorly planned, you will end up with horrible results that completely ruin the experience. The general rule with projection effects as a whole is to have extremely low lighting conditions (something anyone with familiarity with projectors will know), a facial surface adds an additional complexity to the task.
The Buzz Lightyear and Constance figures are examples of extremely poorly done facial projection. Buzz's surrounding environment is a relatively brightly lit queue, so the effect stands out out as a poor mismatch with the rest of the figure and the environment. The surrounding lighting in the environment also ensures the face looks blurry and washed out.
The Constance figure on the other hand has more appropriate lighting conditions at least, but flunks the facial surface. There was no attempt whatsoever to sculpt any of the face texture, it appears it's just a blank flat canvas. It looks totally flat and horrible even head-on, and other angles are just abysmal. It also doesn't help that they decided to animate her arms via projection, which just adds to how flat and awful the figure is. This figure is perhaps even more laughably bad than Buzz Lightyear because there are numerous examples in the same ride of PROPER executions of facial projection, which were created many decades before the current bride figure. Here's the reason the singing marble busts are infinitely better executed than the bride-
The 7DMT figures look fairly decent, with the exception of the Dopey figure. They are decently well lit (within limits of having exterior light leakage) to match the surroundings and the imagineers properly sculpted the facial surfaces accurately to the features of the dwarfs. I haven't ridden Frozen yet, so i'm holding my opinion of it for now (projections don't come across properly via online video).
I have noticed another oddity with the facial projected AA's though- an occasional disconnect between the movements of the bodies and the movements of the projected eyes. I don't know if this is a synchronization thing or something to do with the people animating the figures. Perhaps different people worked on the AA movements and another team did the projected animation. The eye movements don't always naturally match what the body is doing however.
While I don't dislike well-done facially projected AA's, i'll always have a heavy preference and love for fully mechanical faces. The Lincoln face shown at D23 was very impressive. And the new Navi figures teased for Avatar are even more of a step up (with incredible and super realistic motion around the eye area). Mechanical faces can be done improperly as well of course (Ariel), but they can support a much wider range of lighting conditions comparatively speaking and it can be more satisfying to have a fully physical effect. But both types of effects do require very precise conditions to truly pull off naturally.