I'm joining this party a bit late. What's this BYFFI process you're talking about? Since it involves Blender, does it require any special knowledge in order to use it?
When I make a _far.nif, this is the process I use:
* Remove collision from the original in nifskope.
* Export the mesh as a .OBJ file.
* Import the .OBJ file into Milkshape 3D.
* Use Milkshape 3D's "directx tools" menu to bring up the mesh reduction tool.
* Adjust the slider on the triangle count until the mesh quality begins to degrade.
* Export the results back out into a .OBJ file.
* Import the .OBJ file back into nifskope.
* Check it for deformed bits, if things look bad, repeat the work in Milkshape with a lesser amount of triangle reduction.
* Apply low-res textures.
* Save the file, optimize it with PYFFI.
It's a really tedious process, but it has so far served the purpose well enough. Some of the architecture meshes ended up with UV problems but I don't know how to fix those. Most of the problems aren't visible though so it doesn't really matter.
I haven't really looked into/used milkshape 3d much, but that is interesting how much working around you do just to reduce the quality through that process.
And from you're explanation of how you did it, I see nothing wrong with the way you do it. It's just another way to do the meshes.
The same thing can be done in blender as far as degrading a mesh, it just requires knowledge of how to use the modifiers to accomplish this feat.
As far as my "BYFFI" process it involves a bit of thinking on the actual modeling side of the equation. Let me give you an example with one of your meshes.
Ummm. RAEVWD icelvenblock02house03_far.nif 163 kb
Import the mesh (nif) into blender. Open the mesh objects up and delete the duplicate and/or unused vertices. Export (nif). Open in nifskope & delete the NiBinaryExtraData. The simplest way. Mesh will now be about 122 kb. Right click optimize on the mesh itself PYFFI again 128 kb.
EDIT: Another RAEVWD example just by simply deleting unused/duplicate verts: icexteriorwall01_far.nif 298 kb down to 141 kb. Yes, more than 50% easily. 142 kb PYFFI. Time took: 1-2 min lazily.
If you want to take the optimizations to the max, you will have to at least learn the basics of modeling with the program(blender being what I use). This may involve UVmapping also depending on how much more optimized you may want a mesh to be.
If you think about the model(s) logically, one will come to the conclusion that about half of the dome roof object in the mesh is not needed, because it will never be seen or touched or walked upon, because there is no collision. Why? Because it is under the main roof which shows on top of most of the dome portion. Some of the verts/faces underneath can be safely deleted.
Anyhow, the thinking of what to change can go various different ways and is at the discretion of the modeler.
Questionable changes would be dependent on the texture and if you want to make more textures or a specialized texture. Think. Could all the windows on each side of the building be 1 face? instead of 3,4 or 6 or whatever. Does the gutters or roof rafters have to have that many verts or faces? Definitely not. But would require a bit of work to remodel and UV map them, see?... A new reworked texture may/may not be needed to accomplish some of these things.
How many objects can be merged? Do they use the same textures/materials?
Another thing to consider is whether the faces are quads, tris, or a combo of both. This will determine the total # of faces. Simply changing the whole actual model to all quads will a lot of times wreak the UV mapping on certain portions. This is where the modeler will need to know by test what verts to not include.
(Reg model example) On a lot of the models that have stairs leading to a doorway, just look underneath the stairs. How much of that is not needed. Especially in the collision model also. If you understand the lines not to cross within the models, you will quickly understand and add up collectively how much unneeded data/faces/verts there are in the models.
Is it possible to make two collision faces one face? How about six or twelve faces into one? Depending on how many times that particular model is used in an area will effect the benchmark. Why should one have all the extra if it will never be seen, used, touched, or walked upon etc.
In doing so to models this will result in one or both of these: a optimized model and/or extra specific models.
And make sure in this whole process that the normals stay faced correctly. Regular models and collision models.
Some models will be used more than once and in various positions. Will deleting certain portions of a mesh affect the model overall In-game wise?
Above all this whole process is basically thought out as to what is/isn't needed for each particular model.
Most of the problems aren't visible though so it doesn't really matter.
Yes, most of the UV problems may not really matter much, but if you can't see it, then why should it be there?
Learning to UV map is the best way to create new looking models from existing meshes quickly and (sometimes) easily.
Also in correcting UV mapping on existing models.
*End modeling blurb*
Hope this sheds a little light on the whole optimizing processes. Many different paths and choices to make.
Choose wisely. Save a copy of the original in case a portion may need to be added back in at a latter date for one reason or another.
Metallicow