I have a suspicion that the fence pulse might be overloading the Kiwi frontt-end, in which case information is lost anyway before it gets to the noise blanker stage.
Another option, assuming that you have the room at the bottom of your signal spectrum, is to put an attenuator on the Kiwi antenna input and push your received noise floor down closer to the Kiwi's hardware limit. If the numbers work out right it should mean that the peak of the fence pulse doesn't saturate the Kiwi frontend, but still allowing the Kiwi to have the full bit depth at the quiet end to have all of the information decoded from the incoming signal.
If my reading of the flow of wsprdaemon decoding is right, then the presence of the pulse is of no relevance in itself as the protocol will happily work around the presence of the impulse noise; but that makes the assumption that the inbound signal hasn't been clipped or otherwise affected between the pulses due to the impulse noise.
My two KiwiSDRs and the RaspberrySDR that I have pulling signal from an Active-Antenna AAA-1d amplified loop antenna (via one of SV1AFN's "DXer's High-Linearity Pre-Amp (LNA) for HF Receivers" to split and amplify the signal) all appear at times to have overloading, and I am suspecting the ethernet over powerline that the neighbours are claiming not to have. I have yet to try attenuating the signal into the mixer to see the difference, it's a project for over the Christmas.