I pondered a similar idea, but got lost in a circuit design as I would be dealing with a 5-cylinder engine - it would have to allow cylinder activation every other injection pulse (electronic background is touch-and-go hobby level).
Then like other's have said, you have some pumping losses to deal with, and as long as rpm's are high enough (say at least twice idle rpm), than I would think torque pulses from combustion would be similar to what's normally appearent. Additional negative torque pulses would be introduced from dead cylinder compression of an air-only charge I'd imagine...this all leaving the harmonic effects upon the crank and main bearing clearences in question.
I also had concerns about how the OBD2 would react, knowing it likely to trigger injector codes unless given a dummy load rather than an open circuit. Then there would be differing load information it would pick-up relative to throttle position between cut-off activation, as sensed by the mass air sensor, the TP sensor again, and oxygen sensors (perhaps even the coolant temp.).
As this is my one and only ride, I've opted to leave well enough alone. I also do most city driving so I don't think I'd find much of any gain with the system...but conservation is conservation!
In thought, not running cylinder deactivation alternately when engine loading conditions allow (meaning not to have the same cylinders deactivated eash time), on an engine with an even number of cylinders, would creat uneven wear characteristics, as well as build-up potential - oil beyond rings and intake valve seals. That's without considering valve deactivation. With considering valve deactivation I suppose oil contaimination wouldn't be near as much if any concern. I think uneven wear might still be a concern - increased thermal and force loads upon those cylinders that continue firing.
Just my thoughts.