Presumably, a cold tube would leak less than room-temperature (let alone elevated-temperature) silicon?
You can do a sampling ADC sort of thing where the capacitor divider is grounded, zeroed, then switched to the voltage node and sampled. When the sampling is quick, leakage current of the sampling / measuring circuit has little effect. Note that, since this is a charge pump, it still draws current proportional to sample rate and capacitance, so you can't sample very quickly.
1kV and 1pF draws 100nA at a sample rate of 100Hz. Assuming the switching devices have much less than 1pF of capacitance themselves, which is a very tall order (1mA 1500V MOSFETs *could* exist, but I doubt you'll happen to find one available anywhere).
This is why esoteric switching methods seem to be suggested: it's slow enough to use high voltage reed relays; high voltage photodiodes (or phototubes); thermionic tubes (specifically, medium to high voltage, filament cathode types where the thermal time constant is short, the capacitance is small, and the voltage standoff is high); even CdS (or other) photocells, given the appropriate dimensions (it would have to be a very large "bright" resistance indeed, however).
A 1ms time constant (or less) would be appropriate, so the switch can have an on-resistance of under 1Gohm. It should have an off-resistance much greater than this, the difference giving the gain error. I.e., a 1G on, 1T off gives a 1000:1 on-off ratio, which will lose 1/1000th of the signal, or 99.9% gain -- not too bad, so mediocre parts like CdS are suitable here, as long as they are appropriately chosen.
Tim