I'm not actually aware of any CPUs that preform a XOR faster than a SUB. And more importantly, they have identical timings on the 8086, which is where this pattern comes from.