I would certainly agree with THHuxleynew 's post above. The number crunching for a simple charge-discharge cycle isn't difficult, and as long as the resistors comply with an I2R law at all frequencies, then the calorimetry should work (and my crunching actually showed a slight power bias to the input side, too).
However, we know the negative resistance characteristic of the spark gap results in bursts of RF noise (as seen on various oscilloscope photos). The effect of this noise on the heating characteristic of the output resistor is an unknown, as it is effectively ignored in the DC "temperature vs power" calibration method.
I would suggest that we need to devise some extra resistor "temperature vs power" tests which use a range of frequencies, right up to those normally seen during discharge. We don't want to be caught out by any additional RF heating of the resistor - so this either needs eliminating as a possible source of error, or incorporating into the calibration chart.