Separate CTs around the conductors with their secondary currents summed algebraically would theoretically produce the difference figure, but because you are looking for a small difference between large numbers it would require transformers with an order of magnitude greater accuracy than running all cores through one transformer. E.g. a single transformer accurate to 1% with all cores through, will read 1mA leakage in 100A load accurate to 1% of 1mA which is fine. But with separate transformers (say two transformers on a SP circuit) each of which is handling the full load, it would read 1mA in 100A accurate to 2% of 100A, i.e. a possible error of 2000% due to differences in flux leakage, B-H curve, secondary resistance etc between the two. In practice many of the errors would probably cancel but calibration could be tricky.
Much easier to use a single transformer with as low a ratio as possible, with the four conductors nicely positioned in the window so that the flux linkage from each to the core is sensibly equal; run it into a suitable shunt and put a millivoltmeter across it. If the transformer is rated for the full load without saturating, it will be able to transform a considerable amount of power to the secondary under heavy earth fault conditions, worth making sure the shunt will withstand that without opening up and destroying the meter. Power dissipation aside, you can use a fairly high value of secondary resistance compared to a normal CT measuring circuit, as it won't impose a voltage burden on the primary side under normal working conditions. as it would when measuring an unbalanced current.
