That last point has had me puzzled for a couple days Karlos. Yes the Bvision would have a clear advantage in transfer rate over anything tied to an expansion bus.
But I'm still not sure that that alone would be enough to give it an edge in 2D operations.
That depends on what those 2D operations are. If they involve anything software-rendered, such as video or old 3D games or indeed anything that requires the CPU to calculate pixels and then render them to a BitMap, provided the calculation step is not already the slow part and the bus is utilized properly, the BVision will win.
It even wins in some 3D cases. When I first started (back in OS3.x 68K/WarpOS days) on the Permedia2 driver and streamlined the vertex array handling, for large lists of untextured polygons, the bus speed became a dominant factor and on my 25MHz 040+BVision crucified my 28MHz 040+Voodoo3000 in synthetic tests for W3D_DrawArray()
I might still have the benchmark data somewhere.