Most modulation transformers have a gapped core to help it handle the DC current of the RF final. We know that eliminating the DC current, and using a reactor helps a bunch since we get full magnetic BH for the audio only. That said, then we need this AC coupling capacitor, the topic of this thread. But what about eliminating the cap, and just placing the reactor directly across the modulation transformer? If the winding resistances are the same of each winding, then the DC current through the secondary of the modulation transformer is cut in half. Wouldn't this be good enough, eliminate the need for the cap, and eliminate any turn on/off spike?
I recall that Dave's original question was about using an audio output transformer connected backwards to an audio amp originally designed to drive a speaker. If the output transformer was designed for push-pull tubes, it probably doesn't even have a gap, and if single-ended, its DC current rating may be much less than the plate current of the 807, so the coupling capacitor may be a necessity.
The idea of running a reactor without a coupling cap presents some considerations. Firstly, the inductance of the reactor and that of the mod xfmr secondary are in parallel, which lowers the total inductance. The mod xfmr would get worthwhile relief from the DC only if the DC resistance of the reactor were equal to or less than that of the secondary of the transformer. If you have a high-inductance reactor with unusually low DC resistance, then it could be used without a capacitor with no problem.
I would suggest finding a good 2 to 4 mfd non-electrolytic capacitor with a working voltage rating about 25-30% higher than the plate voltage on the 807, and use it. I believe all these concerns about spikes and overshoot, and the need for a DC voltage rating 300%-400% of the actual DC voltage, amounts to a lot of needless worry over a non-existent problem.