Robin, that is an interesting analysis of the AAVSO data.
I’ve taken a brief look at a couple of the papers and there is mention of different durations of totality. For the 1997-98 Graczyk et al (1999) found totality lasted 450d in U-B and 477d in B-V. While for the 1976-78 Saito et al (1980) found totality lasted just 300d, and Saito note this is “shorter than those estimated during earlier eclipses“.
I’ve not read them in detail, though I wonder if this is a clue to what is happening..
The duration Hugh found is closer to 300d than 450d, at 332d using H-beta. This appears similar to the AAVSO data analysed by Robin, where by eye it looks to be just over 300d.