I'm not sure if the BGFX stuff is going to be directly related to the 3D acceleration of drivers.
It's been said that if/when use of the video card is implemented we will be treating the cards like additional CPUs to offload tasks too, essentially doing software rendering on the GPUs instead of the main CPU, they're designed for such complex mathematical tasks afterall.
As already noted, if you simply use the graphics cards as graphic cards (translating game calls into D3D / OGL calls), you end up with inconsistent results between platforms / driver revisions / cards, much like people are finding trying to run the hacked up Taito games with the wrong cards; it doesn't matter if you pass the right commands as the results will always be different and that's no good for us. Very recent cards, and DX11 level tech give us the power and ability to do what we need to do, but we still need the code. Obviously this method is significantly slower, but should still be faster than rendering using the main CPU alone on a good card.
A more logical short term goal /conclusion is that we might be able to do a better internal interface, and have better layout files (for things like the Fruit Machines) with hardware accelerated rendering of those things, not so exciting to hear mind you.