software 3d engines

hello forum, i''m currently working on a software 2d/3d engine, which supports rendering through directDraw or a simple backbuffer copied onto a GDI bitmap. what i''ve got so far is: describing triangles/objects in 3d, transforming (translate,rotate,scale)them with matrix math , simple depth sort (buggy, no Z-Buffer), rasterizing plain triangles in different colors, crappy perspective texture-mapping, all in 16bit along with simple colorkeying for simple transparency effect. right now i''m trying to do some shading/lighting, then i will start to optimize the whole thing, because it really gets too slow when rendering more than 50/60 triangles (i think the code can be optimized a lot!). a few questions: is/was someone else of you working on a software engine? where did you have the most troubles(i have a lot :-) )? i cant figue out a good method for clipping. right now im clipping every single pixel, and test every single pixel against the color key, which i guess slows down the engine very much. why are commercial software engines like Unreal,.. so F***ING FAST, i cant believe that they get 30 fps in 800x600 and still look good :-) bye