that's true for something like a 3-phase external fault (the .58pu on one side is cancelled when comparing to the 1pu on the other side). But 3I0 into a transformer on one side != 3I0 leaving the other side if it's solidly grounded on at least one winding (usually). For an out-of-zone ground fault, you'll end up with weird I0 currents like in the figure I uploaded. Traditional percentage restraint relays will cancel the I0 on a per winding basis (ie, each winding set I0 is cancelled individually, instead of cancelled on a per-phase/element basis). Using that example, the .6pu I0 is removed before the differential compares to the 1.7pu side currents (and vice versa). the I0 is stuck in the delta as dpc said, or the L-L calculation in a modern relay using wye-connected CTs.
