differential mode works when both wires are 'hit' by radiation equally.
if you have one of the members of the pair closer to the signal interference source than the other wire, they are no longer 'hit' equally and you can't subtract out (as well) the common-mode noise component.
I don't think 10/100 cat5 cable has asymmetric twists (not sure) but I know that gig-e (5e or 6) does play that 'unequal distance' game. ie, some pairs end up being slightly longer due to having more twists than other pairs. interesting bit of trivia for layer1 networking folks
(also that is a good reason not to use cat5e or cat6 patch cords for joining things that need to arrive at exactly the same time, such as audio i2s. its somewhat fashionable to use rj45 jacks and ethernet wire to connect high end dac and transports together, and for very short runs, its ok; but its still technically 'wrong').