When we design a part, we do worst case simulations over process variations, temperature, etc.
We also add some margin for tester error. So you have been lucky.
If you choose another AD8666 from a different fab run and test it at -20C or +60C, you may
find that several parameters, such as CMRR, PSRR, gbw, slew rate, etc. do not meet spec.
We have no control over how a customer uses our parts; we can only recommend conditons
that result in a reliable, manufacturable design.
If you have 3.3V available, I would use the ADA4691-2 or the ADA4666-2, both of which are
less expensive and spec'ed at 3V or less.
But I am using AD8666 @ +3.3V single supply in my application.. how is it possible then?