Preamp distortion or power amp distortion. How to tell?

[Enzo]

(What a great song)

[pdf64]

According to that idea (power tubes will be the first to clip), how can we explain then the big difference in the distortion level that exists between a master volume installed in a Plexi previous to the phase splitter or after it with maximum preamp gain?...

With reference to a Marshall 1987 and the master volume (MV) nomenclature described in the trainwreck pages p41-44, the addition of a type 4 MV to a 1987 facilitates 2 cascaded common cathode (CC) gain stages prior to the MV, so assuming a gain of ~60 per CC stage, there's a max available preamp gain of 60 x 60 ~= 3600.
If the type 4 MV is changed to a type 1 or 2, an additional gain stage is available, the LTP phase splitter. This has a gain of ~30, however, there's the insertion loss of the tone stack to factor in, so if that ~halves the signal level, the total increase in available gain will be ~15.
So, there are now 3 cascaded gain stages, total gain ~= 60 x 60 x 15 ~= 54000.
Hence, as you note, a big difference in the available distortion level between the type 4 and type 1 or 2.

...In a Plexi, using a post phase splitter master volume or an attenuator in the speaker output the distortion level is very similar with the gain set to the maximum.

Yes, with a type 1 or 2 MV, we lose the gain of the power tubes, and also the LTP outputs won't hit its own rails until a somewhat higher level than the power tube control grids would clip them at. But on the flipside, the LTP gain is no longer constrained within a negative feedback loop (ie when the type 1 or 2 is turned down, the open loop gain will likely drop to such a level that the closed loop gain will be little different than open loop).
So yes, a 1987 with a type 1 or 2 MV can have a similar level of overdrive as a regular 1987.
However, the dynamic response may be somewhat different, as with a stock 1987 the power tube screen grid current draw will modulate the gain and power output of the power tubes, with that current draw having a non-linear relationship to signal level.
Whereas a 1987 with a type 1 or 2 MV that is turned down will necessarily lose a (perhaps significant) degree of that modulation, as the signal level at the power tubes is lower.