It should be clear that the one-instruction registers on bus processor is much faster than the one-instruction Von Neumann processor. Even for a generous one million bits per processor, a 512 megabyte (over 4 billion bit) registers on bus processor would have 4096 processors instead of just one. Of course, one could make 4096 Von Neumann processors with just over one million bits of memory apiece, but getting them to work easily one one large problem would be difficult. (Shared memory and input/output interconnection could be tried.)
However, how does the registers on bus architecture version compare for a multiple instruction processor? With the registers on bus architecture, the processors can have much less than two million bits of memory. You want as many processors as possible, but they can only be so small because they must probably be no smaller than a bus width (both directions) by a bus width on the chip.
Another thing that has been done to speed processors is to use special purpose circuits. This has been done extensively with some Von Neumann processors, almost certainly including the processor in the computer you are using now.
The processor in the book has just a rotate and mask circuit in the logic unit. However, in a 64-bit computer like the one in the book, one could add a circuit to do 64-bit multiply very quickly (as well as keep the circuit to do rotate and mask). If the program that the computer was running did a lot of 64-bit multiplies, then he program would run a lot faster with the 64-bit hardware multiplier included in the processor.
One can also add special purpose circuits to the register on bus architecure as will be explained.