Perhaps I misspoke. Personally i find it very interesting. Just that I can put little weight on the results in their current state.
Although the total pulse power from the generator is constant the pulse power measured at the core does vary with pulse length. Still, the magnitude of the power compensation is a greater percentage of the pulse power at 100ns than at 300ns. Calculations show that at 300ns the Qreaction is quite small but is of much greater magnitude at 100ns.
It requires more time, and a more complete write-up, to analyse this fully but I find the above sentence worrying.
That is what you'd expect if the pulses generate rectification in the amplifier circuitry, since the pulse power is lower per pulse at 100ns than 300ns. So this is evidence in favour of an RFI artifact - but of course it could have other explanations.