I'm not much of a modeler, but from what I know there's programs and humans that can convert hi-poly hi-rez models into low-rez models pretty autofast. Would it be worth it to make paks of these low-rez resources so popular games redraw faster, look uglier, and work on slower gpus?

I don't think it is as easy as using smaller images and less poly models. Most high udget games will use a custom game engine with very specific needs in the models they make for it. You might be able to make the models smaller and smaller images, but that is not going to stop the shaders, bump mapping, lighting, normals, transparency, ext that is going on in the code.

Now, the opposite has been done:

Some people like to make high resolution textures for old games. Duke Nukem, wolf 3d, and doom have been done I believe.

Would it be worth it to make paks of these low-rez resources so popular games redraw faster, look uglier, and work on slower gpus?

Assuming that works, then whether it's worth it depends on whether you want to play one of those games on a slower GPU.

Back to whether it actually works or not, you'd have to measure whether these games are actually bottle-necked or largely impacted by vertex-fetching, input-assembly, vertex-processing or pixel-block under utilisation. If so, then lowering the polygon count might have a positive impact.

I was always wondering how to make Dead Space iPad version. Are there a fast way to automatically change them into low-resolution or less-polygon models? Will they re-use the code for game logic? Or they have to re-write the source code? Or they plan to do cross-platform development before programming so they develop two editions?