Martin -- I think at this point, it would be good to post some pics of oscilloscope displays that your unit it producing. At this point we'll assume that your 300 is working properly, since apparently the multiplex adapter will still not display separation on a known stereo commercial station, either? Have you checked this since the tuner portion has been realigned? Doing so would serve as a double check to help rule out possible faulty operation of your 300.
For proper separation in the multiplex adapter, there must be the main L+R audio, and the 38 kHz suppressed carrier L-R sidebands appearing at the point where the 1 uF cap, two 100K mixer resistors, and 1 meg ground reference resistor all meet. At this point, the L+R audio simply passes up through the 100K mixer resistors, and on to each grid of the matrix mixer tube. The 38 kHz sideband info passes up through the 100K resistors as well, but encounters the 38 kHz energy applied to the balanced modulators, which detects the suppressed 38 Khz sidebands. This produces an audible L-R signal at the output of one modulator, and an audible -L+R signal at the output of the other modulator, each being applied to one grid of the matrix mixer tube, along with the aforementioned L+R signal to each grid as well.
In the mixer tube, the separation control has no effect of the main L+R signal, since it is an identical in-phase signal applied to both grids. Therefore, whether the cathodes of this tube are shorted together, or completely separated, the degeneration produced across the cathode resistors of this tube remains fixed.
However, the sideband signals (now audible) are identical -- but out of phase -- signals, with one applied to each grid. For these signals, progressively shorting the cathodes together eliminates the degeneration for these signals at the cathodes of the mixer tube, which effectively increases the amplification of the sideband signals relative to that of the main channel L+R signals through this tube. At some setting then, when the levels of the main L+R channel, and the levels of the two out-of-phase side band channels are made equal, then one plate of the matrix mixer tube will produce a 2L signal (the R components cancelling each other out), while the other will produce a 2R signal (with the L components cancelling each other out).
Since you are questioning how and where separation happens, I wanted to detail the whole process to give you maximum advantage. Separation starts a the output of the balanced modulators (because that is where the sideband sub-channels first become audible), and ends at the plates of the mixer tube, because that is the point where the levels of the main and sub channels have been equalized.
The filters at the output of the mixer tube are necessary to diminish the 19 kHz pilot signal, and the useless suppressed carrier 38 kHz sidebands created of the L+R signal applied to the balanced modulators.
The separation process is rather detailed in explanation then, but rather simple in process. Again, I think some scope pics would be most helpful going forward from this point to get to the bottom of the issue.
Dave