Like Elvee wrote, Figure 4 is a normalized graph of the drift of the voltage versus time rather than of the voltage itself: Delta Vz(t)/Delta Vz(infinity) rather than Vz(t)/Vz(infinity).
When you look at the schematic diagram on page 248, there is no component there that could cause a 1 minute time constant - microseconds maybe, but not much more. The only thing that can be quite slow is self-heating, which also explains why the graph of Figure 4 changes somewhat with a small heatsink.
I agree with you that the voltage across glow discharge reference tubes is too high until they start. In fact it is as high as whatever supply voltage you apply to them; feed them from a 225 V rail via a series resistor and you get 225 V until they start.
TV's were not the only type of gear using them: VCR tuners, DVD recorder tuners, FM tuners also had them (there are certainly many other examples)
The ambiguous definition for me relates to the dV/dt of the infinity value. Maybe there are more detailed specs or reports on the drift of the reference level itself, but as I see it, that particular chart is not clear in what information it is trying to convey. Maybe it is related to the test circuit #1 (SS 0038) in that datasheet.
Elvee how did you derive the 3.2mV? (which I see as 0.01% of 33V, not 0.0001%)
Elvee how did you derive the 3.2mV? (which I see as 0.01% of 33V, not 0.0001%)
To add to what Marcel said, the curve shows the effect of the thermal time constant on the thermal drift: for example, if the voltage value when hot is 10mV higher than at 25°C, the graph shows how these 10mV are reached over time
I completely messed that one: 3.2mV is the tempco of 3.2mV/°C, not an absolute value, and I mixed up the decimal places