Exposure works by multiplying RGB input by a constant equal to 2^{EV}, so for example, adding 2 EV results in a multiplication by 4, subtracting 2 EV results in a multiplication by 0.25.
Zone system works in Lab, scales the L channel logarithmically then split it in zones, which really does not make sense at all (EV readings from Lab are not EV). The next drawback of this module is you can’t push a zone without pulling another. The last but not least is, as usual, doing magic on the L channel messes up colors in a strange way.
I’m working on a tone equalizer that uses the same algorithm as the exposure but allows to define a different exposure compensation for each zone, then blends them using a gaussian series. It’s WIP, see there :