If I may, there is a fundamental point that needs to be understood throughout this discussion: The application of Global Negative Feedback (GNFB) does not change the gain, output capability, or performance of the EF86 stage in any way -- nor any of the other stages within the GNFB loop for that matter, either. The open loop gain of the design -- as determined by the total gain of the various stages included within the loop -- remains unchanged whether the GNFB loop is open, or closed.
The NFB voltage is simply a "counter" if you will to the signal voltage applied to the stage that the NFB voltage is applied to. This effectively then reduces the signal voltage applied to that stage (by the amount of NFB applied) -- which it then amplifies with the same gain characteristics as it would were no NFB applied.
The salient point then is that when GNFB is applied, more signal drive is required into the stage it is applied to, versus the level if the feedback signal were not present -- not because the feedback has affected the gain of that stage, but because the effective signal applied to it has been reduced. What this really does then is require more of the preceding stage (or equipment) driving the stage where the NFB signal is inserted, because more signal is now required from that stage or equipment -- but nothing changed with the stage where the NFB was inserted, as has been suggested here.
If the EF86 stage does not show clipping at its output when maximum power output is being developed with the NFB loop open, but does when it is closed while producing the same power output level, then either the increased signal level now required to be presented to that stage is now being compromised, or much more likely, what you are really seeing is the beginning effects of a clipped output signal (produced no doubt by the output stage) being applied to that stage through the loop. The effects of that signal will be mixed with the input signal supplied to that stage, with the composite of those two signals produced at its plate in the same way that an "undistorted" NFB signal at lower power levels mixes with the input signal as well. A determination as to whether the stage where the NFB signal is applied is clipping or not under full power conditions can only be assessed then with the GNFB loop open.
Now, there is one corollary I will add to this discussion: If the stage where the NFB is applied is biased such that it requires the NFB voltage applied to it to keeep it from being overdriven when full power output is produced (i.e., where static bias voltage + peak NFB voltage is equal to but not greater than the peak input signal voltage required into that stage to produce a full power output conditon), then that produces a condition where the input stage (where NFB is applied) can in fact clip at the moment the output stage clips, because once that happens, then there is no more peak NFB voltage capable of being supplied to that stage to counter any increase in input signal applied there. The result is that the input stage instantly clips when maximum power output is reached.
This is why it is important to ensure that where any GNFB signal is inserted, there is enough static bias voltage present to ensure that the stage won't be overdirven at or shortly after maximum power output hs been reached. Having enough open loop gain (OLG) within the GNFB loop is the best way to ensure that this won't be a problem.
You might do the math regarding the bias conditions of the EF86 stage when maximum power is being approached and in fact reached to ensure that this last scenario is not what's occurring in the Allen design. Most capable designers ensured that this problem wouldn't occur, but not everyone abided by it -- when I did my rework of the Bogen MO-200 many years ago, I finally found out why the sound level from all those Bogen MX-60As I heard back in my band days in the 60s would pump and fade in and out as maximum power output was reached in those units when operating under what I call "band conditions": Yep, the input stage of the power amp section where the GNFB was applied would clip instantly when the output stage had no more "bias voltage" to apply back to the input stage. Not sure if Bogen ever learned that lesson or not, as pretty much all of their 8417 designs included that same flawed input stage design.
I hope this helps!
Dave
