It would be good to see the complete distortion spectrum - I had a play with something similar and feedback to the resistor did indeed reduce H2, but increased H3 to a level higher than the original H2, so it may be more complicated than it first appears.
It is true that H3 increased a small amount, the overall distortion went down significantly. Not sure about the origin of H3, it may be dominated by the output stage we are feeding back to the input pair. A study for another day. Overall, it's a reasonable technique to stay true to an all tube approach the old timers were limited to. A tube current source could be used but requires another bottle with the cost and complexity. No solution is perfect but I believe this is a good one.
Fender 5F6a amplifier has a connection from output into the tail of the LTP driver. It has been W*I*D*E*L*Y copied, less widely speculated.
Side note: much of 5F6a seems to be "adopted" from Gibson Vanguard 77. Tone control and even a Presence in the output stage NFB. Yet everybody copies the 5F6a. The 77 is cathodyne and the 5F6a is the then-novel LTP, plus the NFB into the tail. With Presence it is not only a tone control but may be a distortion flavor control, changing the THD vs Freq curve.
Side note: much of 5F6a seems to be "adopted" from Gibson Vanguard 77. Tone control and even a Presence in the output stage NFB. Yet everybody copies the 5F6a. The 77 is cathodyne and the 5F6a is the then-novel LTP, plus the NFB into the tail. With Presence it is not only a tone control but may be a distortion flavor control, changing the THD vs Freq curve.
Thanks PPR, nice to know the history. Independently arrived at, but clearly there are many that beat me to it!