Unfortunately it doesn't work like that in real life.
Besides the guy hasbarely defined muscles, this isn't like some of the guys in the latest tekken which sport real musculatures.
Polygons are not always evenly distributed, its in the hand of the artist. Remember at the early stages MGS4 Snake was said to have like 60k vertices or something high just for the hair alone. It was later reduced by one significant figure, if I am not wrong.
Snakes beard alone is meant to contain as many polygons/as much detail as a enemy in MGS3. So unless the PS2 was pushing multiple 60k characters on screen, I'd say you got your numbers confused.
So I don't know why (beside budget) for realtime closeup cinematic they don't used 300k poly models, I am sure the GPU can handle them.
I couldn't find an untextured fenix, so that's what I used.That is not a Gears of War ingame model.
That is a model that's been used for the Unreal Engine 3 presentation, back when it was revealed, sometime in 2003 or so. The actual ingame Locust character is quite different looking, having been reworked a few times; and it has more polygons as well.
And this is not the Marcus Fenix character anyway.
What does a Tekken character got to do with either Gears or KZ2? It's irrelevant to this discussion.
Polygon counts are defined by one and only one thing: what the programmers set as upper limits for the artists. If the budget is 50K, the artist has more freedom; if it's 10K, he'll be more conservative. That's all there is to it.
Edit: just so you know, I've been working professionaly as a 3D character modeler for the past 8 years... I'm talking from actual experience here.
That's a character design issue. El Blaze is a speed character, not a power character, so that makes sense. Try a power character, like Akira for something like that.
My point is that it would likely require more polygons for a muscled character with form-fitting clothing to be properly represented than for an armored one, as it would more easily give off the presence of a low polygon count.
Maybe becouse of collition detection and how such high numbers for a model also means drastically increase in overall polygon count when it cast volumetric shadow(s)?
For cinematic closeup, you don't need collision detection. If shadow is a problem, it can be pre calculated what ever you need for cut-scenes and stored on disc.