Well, so far the tally's pretty much split evenly with 3 for precision, 3 for speed, & 6 for a mixture of both (which really doesn't apply in the matter of the reason I was asking). So, I decided to go with precision & hope that faster processors help to offset the speed difference. Slower processors are just out of luck. Renderable data will be implemented as 64-bit floating point values.
In other news, I now have a pretty good layout for the pipeline. I'm not sure if I can accurately describe it in words without pictures & diagrams, but I'll try:
Functions that directly change the context will do so immediately. Colors, vertices, normals & things of that nature will be contained in the GL's context until it's time for processing. Processing won't begin until GLEnd() is called. At that point, the context & all of it's gathered data will be sent through the pipeline. All the pipeline does is render the graphics as specified by the data within the context. the only 3D functionality exported by the hardware driver is the entrance to the pipeline function. If there's no hardware pipeline function attached, the pipeline will use it's internal software pipeline.
That's all for now. If there's anyone here with 3D experience & a few comments, tips, or anything helpful to share, now's the time to speak up!