Quote - Hm, but you normally cannot the determine the speed of the processor by measuring the speed of a single instruction.

Yes you can, first you can know the intruction speed and latencies for each instruction in the data manuals for your processor.

You also can measure the execution speed in CPU clocks for an instruction or group of instructions, this is the way how codes are optimized.

Quote - If 64- and 32-bit processors were the same in every aspect, except for the number of their registers, the 64 would still be faster on average (for any real program).

64 bit and 32 bit instructions run at the same speed in most cases. 64 bit can be a bit slower because require an extra byte prefix for the code, also multiplication and division is slower.

In most of a 64 bit code you use 32 bit instructions, you win nothing using 64 bits. If you want to count from one to ten you don't need a 64 bit register, even a 8 bit register is enough. Much more, in most part of any 32 or 64 bit code 16 bits are just enough, you only use 32 bit because the processor is optimized for 32 bits and run a bit faster than using 16 bit registers.

You only use and muste use 64 bit instruction for pointers to memory access, it work with the same speed as 32 bit pointers but you are able to access memory in any memory area and not restricted to the 2/3 GB range.

64 bit code have some advantages that can make run faster the app if you optimize the code because you have eight 64 bit registers extra and eight 128 bit SSE registers extra, you can use this registers for processing instead of accessing memory.

Stupidity also evolves!