Irrlicht Engine

Its slow maybe because its rendering 4 million primitives each frame, with pixel lighting and soft shadows.

that sounds like it could be optimized pretty easily (4M is almost always possible to reduce drastically)

Its already fully optimised to the point quality is lost... and the meshes are all up in the GPU for rendering.

oh well, then back to my initial recommendation - profile the code, see what takes a long time (if I'd wager a guess: soft shadows or transparency) and see if it can be made faster.

4M tris in itself isn't necessarily a bad thing, I push significantly more on my rather ancient HD5770 so blech.

What takes time is for the GPU to do the rendering of 4 or 5 million primitives!

I don't think its a problem in the code, simply the volume of primitives for the GPU to handle, and run the pixel shader.

Its also not my pixel shader, as if I select the default renderer, its not much faster.

The shadow map takes almost as long to render as the scene, which is to be expected as its usually a similar number of visible primitives.

The delay is in the render call of the GPU-preloaded mesh.

Have you ever tried to render 4 million primitives and seen that the GPU takes time to do that?

Render a model/s with 4 million primitives with your code and tell me what frame rate you get, and on what CPU/GPU!