The plot thickens.
The most recent observation is that there is a HUGE variation in the inductance of the toroids depending on the turns AND the spacing. I tested this because my first attempt at winding the "better" filter toroids produced unexpected measurements with my LC meter. I then remembered the QCX LPF tuning that Hans Summers has documented for the QCX. I now have another test to perform. Now I need to see if adjusting the spacing on L1 to L4 can optimize the LPF.
Based on my LC meter, the inductance of 9 turns on a T30-6 can vary from 290nH to 310nH. It could mean that "tuning" the inductors could balance out the power levels. That could also explain the production variations, coupled with the gain differences in Q90.
Oh well, more tests to document.