Thanks guys! Im still working on speeding up a few routines which should hopefully improve the speed a little more.
As far as grayscale goes im thinking 3lvl would be possible enough. I think it might suffer too much if it was to use interrupt driven grayscale however. For a cube though it should be possible

.
EDIT - I rewrote a fair amount of code so that objects can now exist and be transformed within a 16-bit world. The good news is that the performance hit is around 2-3fps and i havent attempted to optimise yet:
Small test, zooming in from afar:
EDIT 2 - I have recreated the low detail (most are 3 polygons!) star wars models i had earlier (well most of them). Here you can see: xwing, awing, ywing, rebel transport, rebel frigate, tie fighter, ti interceptor, imperial transport, star destroyer:
EDIT 3 - I have implemented some basic level-of-detail which will drop polygons from a model (and turn off backface culling in some circumstances) when an object is a certain distance. The amount of polygons to reduce each hurdle (there are 2 hurdles) is defined on a per model basis. The hurdles are set lower than normal for testing but the performance increase can be quite significant:
EDIT 4 - So many edits! It seems that no-one is replying

. At any rate i am still working on this project. I made contact with qarnos and he was able to provide me with a GREAT deal of information on knowledge on how to correctly combine rotations as well as helping me with some other problems and as such i have correctly (i think) implemented my camera controls as well as proper movement in 16-bit 3D space. Here is a screenie of 9 cubes (108 triangles) running @ ~20fps with level-of-detail running:
EDIT 5 - I have made a utility to convert 3D models in Wavefront OBJ format to data useable by this engine (including which pattern each triangle should be). It requires a couple of things to be set up in Blender (The modeller i tested with) but it makes life a LOT easier

. Here is a rough sphere test (80 polygons). Note that i have a lot of optimising to do with this engine so 80 polygons is a bit of a hit!:
EDIT 6 - It's a me, Mario!
EDIT 7 - Dark Link!
EDIT 8 - For those interested, I have found an issue with the engine and rectifying it should give me a performance increase ... stay tuned!
EDIT 8 (cont) - I gained a few fps doing some restructuring. I have also turned on depth sorting which is a bit of a performance hit. Still the overall result is still faster than the previous screenies:
