I agree that this does make a lot of sense. It has been also discussed in other threads before, see this for example. My comment back then was that we should think whether this kind of change makes sense to combine with the openhab2 changes (splitting the modbus binding to transport bundle etc.). @rossko57 had also good comments regarding error handling with individual polling in place (e.g. this comment).
I must comment that pretty high polling rate is already possible with the 1.9.0 changes, @mbs38 managed to achieve ~1700 updates/minute using serial modbus slave. With the introduction of updateunchangeditems, the CPU usage remains low (issue previously due to the mere quantity of openhab events in the event bus). Naturally you are right and in a way we are wasting resources if some the slaves change less often…
Best,
Sami