The originators and promoters of ABX (e.g., Dave Clark, Tom Nousaine, Peter Aczel, etc.) had no problem with the comparisons going over long periods of time, like months. Anyone who feels "pressured" or "hurried" is not taking a valid test.
When the test remains blind and controlled it can be stretched over a long time.
This is very different from sighted comparisons over a long period.
Anecdotal: I one built a pre-amp with 2 circuits I could switch between. One DC coupled with OP227 and compound decoupling and another one using the TL072 with no power decoupling and a bunch of non-biased electrolytics in the signal path. Using a toggle switch at the rear of the pre-amp (so not visible)
I always could tell what was used and demonstrated the obvious differences to those that were interested. They all heard it too.
Then, one day I forgot it was in the 'poor' position and listened to it for months believing it was in the 'best' position.
I heard all of the 'goodness' in that period and none of the 'poor sound' I could so easily tell when switching. The 'poor' circuit was not that poor.
Also had a girlfriend that I instructed to switch cables every now and then when I was at work and keep log. Alas she only did this one time over the month yet I thought this all went as planned. My 'notes' on sound were worthless in the end as I thought I heard differences where there were none.
So... long term test says nothing you can hear differences where there are none but expect them to be.
The discussed test, however, has nothing to do with long term memory nor is that more 'accurate'. The test also was not immediate. No idea how smart the ABX box is (monitoring levels) that might be used for part trois.
If he can still do it then an investigation should be started as to the why.