"I really have trouble with all those new per-pixel and whatever stuff, I wish I would understand all this ;-("

Go to the developer section at www.nVIDIA.com and download some whitepapers (gl extensions) + all powerpoint presentations about register combiners and the demos (per pixel lighting/bumpmapping).

You should also download an app called "ShaderAid" (at nVIDIA of course) so you can understand it better.

Another good tip is to use VAR''s instead of CVA''s (or both?) because it''s a little slower with all those pixel effects.
(minimum of 1 multitex pass + 1 decal pass for Diffuse only bumpmapping,
minimum of 2 multitex passes + 1 decal pass for Diffuse+Specular bumpmapping )

Hope this helps...

I think you could optimize the water code you were having problems with earlier in this thread by making the memory access more linear. Reorder the reads so they are linear in memory (remembering that C 2d arrays are row-major) and you should improve your cache hit ratio. Since RAM is usually the slowest part of the system, optimizing for cache use is usually much more effective than optimizing for math. You could also try aligning your data structures to 16 byte (for vertices/vectors) and 32 byte (for larger stuff) boundaries. That will make cache performance even better.

Could i grab a link to the whole source file, im very interested in this engine after all the praise its got.