The reason why the PentiumPro sucked with 16-bit software is rather simple. When the PPro was on the drawing board, its designers had consulted with Microsoft. MS had said not to bother with any 16-bit optimizations, it's going away. Intel is like that's great and all. They used the die space for something else.
When the PPro was ready to ship, Microsoft was like, oops! We couldn't entirely eliminate 16-bit code just yet.
So later when PII was made, Intel put the previously removed 16-bit optimizations back into the chip.