Well, the comparison between those three models is actually an interesting question, if we don’t start with a conclusion… The following are just my own thoughts, in no way the definitive stuff on any of these models or the thoughts of the people behind them.
The Gollum mesh was fitted to the sculpted concept maquette from (I think) Jamie Beswarick, based on the previous version and a sketch from Christian Rivers. The edges follow the facial features both in the neutral pose and in all the possible extreme poses of stretching and compression, in order to accommodate all deformation using only blendshapes.
The Neytiri mesh is - most likely - fitted on a concept sculpt, either digital or digitized, and the edges are based on all the possible extreme poses again. All the dynamic facial wrinkles are created using just the vertices in this model, probably sculpted in Mudbox, and the actual rig is once again based on the FACS action units and corrective shapes to fix overly distorted combinations.
The two main differences compared to Gollum are that, first, today’s computers are 15-20 times (or more) faster then what we had in 2001-2002; and, second, that Neytiri is a young, beautiful female, with tight, smooth facial skin, whereas Gollum is an old, emaciated person with sagging skin, that has permanent folds and wrinkles even in the neutral pose. I strongly believe that both the old na’vi characters in Avatar, and the more extreme faces in the upcoming Tin Tin movie, will be more similar to Gollum then Neytiri, although with a much higher polygon count, thanks to the faster computers of today.
Clu’s model in Tron is something we know less about, but from what I’ve been able to gather, the rig works like this.
DD has built all the individual basic expressions in Mudbox as multi-level sculpts, down to the pore level details. These are based on their scans of the current day Jeff Bridges, provided by MOVA , but obviously altered a lot to make him look 35 years old, using various photographic references and at least one sculpt by Rick Backer.
They have all the shapes exported from not just the lowest level, but from several others, so the same face rig in Maya can load at least two, maybe three different levels of geometry detail, depending on what the animator wants for his viewport playback. If you look at the DD behind the scenes movie, you’ll see at least two different versions of Clu’s head in the wireframe overlay shots.
The final renders use displacements for most of the wrinkles and folds, except the nasolabial (smile/sneer fold) which are created from a large set of expression specific texture maps. Since they have more than a hundred different blendshapes for their expressions, I guess this means that they also have more than a hundred different displacement maps. For every single rendered frame, a Nuke compositing tree is calculated to generate the proper, frame-specific final displacement by blending these individual expression specific displacements.
(I can dig up some links to support the above tomorrow, at work, if necessary)
So, the mesh is more generic, because it does not have to accommodate all the deformation, a lot of that is only represented in the wide set of displacement maps. I also think that they are not using corrective shapes to fix A+B combinations that are problematic, but rely on some procedural tools to redistribute the polygons over the surface to kind of simulate the skin’s behavior (and considering that they have 100+ shapes, a lot more then FACS, they may cover the more extreme A+B combinations with some of those). This is a different method from what Weta has used on both Gollum (except for the wrinkles on the bridge of the nose, which used a single animated displacement map) and on the Avatar characters, which relied on just the base mesh for everything. I also believe that this is the same approach that they’ve used on Benjamin Button.
What I would really like to learn more about are ILM’s facial models and rigs used on the house elves in Harry Potter (assuming that they did Dobby in part 7), the creatures in the Spiderwick Chronicles, and of course Davy Jones in POTC. We do have a long thread here on CGTalk covering Davy in which the modelers indicate that they were using blend shapes as well, but I’m not sure if it was FACS based. I’ve also seen a short movie on youtube that had the Davy face in zbrush; you can’t see the wireframe but you can still pick up some of the poly faces there.
I also have a short but nice video of Gollum talking which I’ve uploaded to youtube.
You can see how the edges move around as a result of the base FACS shapes and the corrective shapes fixing the (probably) messed up combinations. Goes a long way to explain the topology, I guess.