I too have 2 pairs of 800's. My 'early' pair (s/n 52xx) and older pair (s/n 40K).
And yes the Senn supplied frequency graphs are different.
And yes, I too found differences in their construction both of the drivers themselves as well as the rest of the structure.
Still the audible differences when stock were not all that great, despite the Senn. frequency graphs (which are nearly worthless IMO).
Which leads me to the topic of unit to unit variability, regardless of manufacture date.
And statistically speaking a sample of one early and one later production examples is insufficient to claim that this variation applies to all of these early vs later production examples.
If it were possible to measure 'many' pairs of 800's with a scattered distribution across the entire production run, these unit to unit variations would become clearer, and perhaps a true picture of any production run changes would emerge.
And then again, perhaps not.
And the oft claimed frequency aberrations in the 6KHz to 8KHz portion of the frequency response is not the primary culprit for their 'harshness', at least not in my opinion, which is based upon my years of fussing with these HP's.
It certainly is easy to assume such, what with the peak in that band sticking out over the rest of the FR graph.
But I have found that the single most egregious contributor to this 'Listener Fatigue' issue is the amount of leading edge overshoot that is purposely designed into them.
I have termed this the "Sales Enhancement Tool" which was used to 'help' sell them.
And this "Sales Enhancement Tool" is widely used by most HP manufacturers, and in my opinion much to their as well as our detriment.
And it was the SAA mods that 'clued' me onto this issue of Listener Fatigue and it's sources, in the 1st place.
Just my 2¢ worth of observations.
JJ