So what I'd like to know is other stuff like that.
memory pools of shared assets.
IE lists of meshes, textures, models, and animations stored or paged into memory.
for rigid body animation (non-organic, static mesh limbs, no skin weights), a model can use any mesh and any texture, and any animation designed for a model with similar limb layout. animations (such as a generic quadraped walk animation) for smaller models will probably need to playback faster than for similar larger models. this is assuming forward kinematics as the animation method used. sharing of IK animations for rigid bodies of similar limb layout is something i've never messed with. i seem to be blessed with a wee bit of animation talent, so FK works fine for me. it also allows the artistic exaggeration of gestures, poses, expressions, etc that really bring an animation to life. something IK doesn't do - it just puts bone A at location B.
mesh morphing for facial animations, lip syncing animations, editable character faces, all that stuff - very cool, but technically just eye candy, pretty much zero gameplay impact. if you don't have the dev resources you might consider foregoing such things. same for multiple or custom outfits or gear. just more models, animations, meshes, and textures you have to make that add zero gameplay.
makes it look better? yes. but i have a saying:
"Pictures are for looking at, games are for playing!"
depending on method of implementation, shadows are another form of eye candy that can be cut from the game to reduce the need to create assets.
same for almost every other whiz bang effect you've ever heard about. basically "<anything>map", light maps, bump maps, etc. IE extra data (usually texture type data) used to render a mesh. all you really need is perspective correct texture mapping with basic gourard and phong lighting - and maybe auto-generated mip levels for a 1st person view game.
never underestimate the use of tools to automate the process. auto triangle reduction can be used to quickly create multiple LOD meshes from a single high rez mesh, for example. always look for a tool first (or make an in-house one), and do it by hand as a last resort. that being said, you have to break a few eggs to make a real mayonnaise, so unless you can rely 100% on existing assets you can edit, you will be doing some stuff from scratch by hand no matter what - like me having to make a 3d model of a ww1 flak gun for AIRSHIPS!.
for example, i use a batch file to copy concatenate animations onto models to create the .x files used to load skinned meshes into the game. so i just need to make one female attack animation for example and export it, and then the batch file automatically adds it to all the female models used by the game.
for skinned meshes, texture and animation sharing is possible. as is sharing of rigged clothing between identical body models. animation sharing may require identical skeletons. i've had issues with sharing animations between similar skeletons at different scales not working. but that may be due to the way they are animated, and thus may be specific to the modeler (blender) and skinned mesh library (dx9 tiny.cpp sample code based) i'm using.
for multipart skinned meshes, different parts can be shared (interchangeable heads and bodies for example). note that seams between parts can be an issue. i was unable to use interchangeable heads in caveman due to neck seams not always being hidden by clothing. even co-planar faces at the seams doesn't work, as the lighting does not lerp across the seams, since they are two different meshes.
an example:
Caveman 3.0 uses (so far, its still under construction and final graphics are not done yet):
~280 meshes
~450 textures (256x256, adjusting mip levels prevents blur at close range in a FPSRPG).
~170 rigid body models
~165 rigid body animations
10 skinned mesh models
38 skinned mesh animations (with 1-2 dozen more still required)
~210 wavs
these implement:
a 2500x2500 mile game world with all types of terrain.
~50 types of monsters
~400 unique female characters (different faces, skin tones, hair style, hair color, eye color)
~400 unique male characters (different faces, skin tones, hair style, hair color, eye color)
~240 different types of objects in the game including about 60 types of weapons and 30 types of armor.
lighting is sun or moon + any campfires or torches. directional light for sun/moon, point lights for campfires and torches. DX9 fixed function pipeline lighting (IE basic gourard and phong).
mipmaps are automatically generated by DX on load. as i recall, some mip level adjustment is done at draw time for some objects (like big rocks you can walk right up to).
little or no special effects. gourard and phong are provided by the dx9 fixed function lighting engine - just add a light and make sure your meshes have normals. alpha test for sprite textures, alpha blend for clouds, and a simple 2 stage texture blend for snow. all quite doable with fixed function. in fact i was surprised to find only a 6fps difference between the slowest fixed function method and the fastest possible skinning methods (like 172 vs 178 fps on my PC).
using photos for textures instead of drawing them yourself is also a great time saver.
needless to say, always try to use existing assets first. so for example, right now i'm searching (unsuccessfully) for a public domain 3d model of a WW1 flak gun for AIRSHIPS!
and of course the best thing you can do is improved your artwork skills. the first animated mesh may take a week or two, but by the twentieth or so, that's down to half a day or a few hours.
but based on your questions and samples posted, i'd say your problem is with your obsession with "massaging every vertex until its perfect". i understand totally. its easy to become obsessed with perfection in 3d modeling. i must make a conscious effort to not fall into it myself when modeling. and extremely low poly will just make things worse, as you have fewer vertices at your disposal to get the correct shape.
you may also want to revisit your pre-conceived notion that low poly is a must. its only a "must have" due to the limitations of the artist (takes them a long time to make high poly), not the hardware. in caveman, 20K+ tris is a low poly skinned mesh model.
so i'd say your bottleneck is in hand creating low poly models and hand painting textures, instead of using existing in-house, PD, or 3rd party models as a start, and editing from there, and using photos to make textures, and using in-house, PD, or 3rd party textures where possible. with caveman, so far i've been able to get away without having to buy any 3rd party assets. either i find something PD i can work with, or "roll-my-own" from scratch.
so in the long run, the biggest time savers are use of existing assets (in-house, PD, or 3rd party) as a start, and designing the game with maximum asset re-use in mind.
but until you get over the urge to tweak vertices forever, you'll never get anywhere.
try setting yourself a deadline. ok, i have 1 day to do this model - after that, no more tweaking
once you get better at it, cut it down to 1/2 day per model, and so on.
once you learn to recognize when you're wasting time tweaking vertices, and learn to force yourself to stop obsessing over every vertex, your production rate will improve greatly (probably double at least) - mine sure did. <g>.
the sad thing is i can just picture you spending days and days modeling and texturing the examples you posted, when in the same amount of time you probably could have gotten much closer to AAA results by editing existing PD or cheap 3rd party assets. been there, done that.
not that finding existing assets is easy. especially free ones of high quality. but for common models its possible.