The zener changes the nature of the feedback signal. It is the result of a subtraction rather than a division. This enlarges output errors as seen by the FB input.
To see how, consider this. If Vout is 0.25 V above the setpoint, the FB pin sees the entire 0.25 V error. OTOH if you replace D2 with a 5.25K resistance (the required value to boost 1.2 V to 7.5 V), the 0.25 V error appears as only 40 mV. Not a problem, as the chip is designed for this.
However, the zener approach might require different loop compensation. The zener approach appears to the chip as a weird external gain stage inside the feedback loop, something the compensation scheme (the Vc pin components) almost certainly are not designed for.
Another problem with this approach is that now the output voltage accuracy depends on the error bands of two independent voltage references. The 1618 internal reference is spec'd at +/-1%, but the zener has a B suffix that usually means +/-5%. That is a large difference; I would look long and hard at the zener option before deploying it.
