Our testing show that if you calibrate at a certain temperature the PLL will stay in lock if the temperature changes by less than +/-40C.
Since the VCO uses an internal voltage regulator, the frequency drift over supply voltage is negligible, at least above ca. 1.9 V unregulated supply voltage. The charge pump runs on un-regulated voltage and the charge pump current might change the PLL BW with significant changes in supply voltage resulting in a non-optimum PLL BW (i.e the phase margin might get low). We have not measured the maximum drop, so to play it safe I would say the maximum drop as 1.0 V.
I have also come across a case where the CC1101 was calibrated and the PLL in lock. However, when strobing CSn there was a coupling of the CSn signal to the crystal oscillator resulting in the PLL going out of lock for a certain period of time (don't have details on timing - sorry). Simple test is to cut the CSn trace at both ends and running a jumper wire to check if the problem goes away completely.