You might be right. But philosophically speaking there would be very little return to the designers to optimise nop decoding. I remember that AMD went to great lengths in their first AMD64 manual to explain that nop was now a "true" nop that did not update EAX (and thus would not zero the high order RAX bits). That said to me that x86-64 forced them to start decoding 0x90 as a true nop. Of course Intel may have followed a different path and decided to optimise nop earlier.

But I've yet to see any properly optimised code that uses nops in the critical loop, so optimisation of nop is kind of wasted effort for the CPU makers. And indeed some programs actually want nop to be unoptimised and use up time so as to create a delay.

a standardised delay nop can be a fun thing.

for exemple, if intel decide to create a time instruction.

for example:
1 nop = 50 nanoseconds 0.01%

or somethin else.
and then, no matter the time needed by others parts of code. this 50 nanosecond delay (20MHz fixed clock) will be very fun for many purposes.

and if nop can use rep prefix? lol.
and rep can use an argument (operand)

"rep nop" is now the "pause" instruction.

Modern NOP only consumes decode resources of the processor - it must be optimized in some way for that to happen. Even in the latest processors NOP is used for timing. Albeit rare cases where it is needed.

its nice of you all to talk so much about me but i believe i have some more uses that nobody has really mentioned yet