Even without superscalar execution it'd still help, since no sane architecture has floating-point arithmetic with an effective latency of 1 cycle. Current Intel

Even without superscalar execution it'd still help, since no sane architecture has floating-point arithmetic with an effective latency of 1 cycle. Current Intel cores have a latency of 3 cycles for FP add, and 4 cycles for FP multiply. OoOE is another matter though...