Unfortunately, the RAC Failure Mode Distributions do not show any separate data for glow lamps. The are semingly thrown together with the incandescent annunciator lamps, which makes the data not applicable. (Open, short, and anything related to filaments...)
All sources give higher lifetime for 220V indicator neons than for 110V types. Behind this is the bigger value and factor of influence of the limiting resistor on the operating current, which is very small when 110V and coated cathodes are involved.
O.P.Herrnkind in 'Glimmroehren und Kaltkathoden-Relaisroehren' states that electrode sputtering 'Elektrodenzerstaeubung' is a certain factor, which is proportional to the operating current. He also states that it is cathode sputtering which is causing the dark deposit. Further he states that the end of the lifetime of any glow-discharge devices is characterized by steadily raising ignition and maintaining voltages, with erratic changes also occuring. It is also noted that a lower-than-intended cathode current should be avoided too when the electrodes are oxide coated. The reason is that if the current is too low to keep the complete electrode area involved in the discharge, the non-emitting spots will passivate over time and not enter emission even if the current is increased again.
Mulder in his book 'Gas Discharge Tubes' (1963) states, that :
1.) cathode sputtering is dependent on current and gas presure, with higher pressure reducing the sputtering.
2.) cathode sputtering is the prevalent, but not the only effect which is determinating the lifetime. He names the 'cleanout'-effect, which he describes as being both physical and (to a lesser degree) chemical in nature, which causes the density of the gas filling to diminish. The first phase of this is happening, when ions encounter surface defects on the anode and/or the envelope. the second phase is characterized by sputtered deposits trapping additional gas molecules when forming the abovementioned deposit.
3) that SOME sputtering (during production?) is beneficient due to a getter effect, which binds non-noble gas molecules.
4.) that the envelopes, which feed a metal wire through glass without special considerations (Kovar etc. as well as design) are not to be regarded as reliably hermetically sealed.
This last point is exactly what I would have suspected to be the secondary failure mechanism, i.e. leakage and impurities, as thermal expansion and contraction will open ways for molecules to get out of or in to the tube's envelope.