Update:
It is really looking like a non-starter for now .. Sorry everyone.
I have just run several full simulations - this is the entire EW pitch and VCA / tone sections .. The full schematics were entered, including 'real' oscillators (including equalization coils and a simulated player to test effects on linearity), diode mixer, LM13700 and all that precedes and follows these.. Did not include volume circuit - simply biased the OTA to achieve variable gain.. Likewise with the tone controls, which I biased and tested at extremes and mid positions.
This is a simulation - so one cannot fully rely on the results, but..
1.) The FM/PM distortion due to change in capacitance on the diode mixer, has almost certainly been ruled out as the cause of the 'problem' .. It is quite difficult to be absolutely sure because of its nature and the way the simulator operates in 'timesteps' - but I ran an extremely high precision simulation to verify this (My PC was tied up on this for more than an hour) but found nothing significant.
2.) Oscillator synchronisation is the primary source of deviation from sine on the input signal (signal across C23) - and it is quite severe at low frequencies (I was able to adjust my players capacitance down to get 110Hz - but each simulation takes so long I didnt try to go lower) - there was severe distortion down at this frequency.. This distortion extended up to 1kHz - the distortion decreasing proportional to frequency until, at about 2kHz, a pure sine wave appears across C23. The filter (C23) is set quite high (>10kHz) so the reduction in distortion is not due to the filtering effect of this.
3.) The distortion on the input waveform may NOT be the primary problem - but while it is there, there is no possibility of getting a pure sine wave below about 1kHz .. One can reduce the distortion by increasing the capacitance of C23 and C26.
4.) Thierry is right about the VCA / Distortion stage being the most likely "trouble spot" .. Distortion produced by this can be really horrible - I only ran a few FFT's, but these showed that, although the input waveform is distorted, this distortion mainly consists of musically pleasing harmonics.. Feed this waveform into the VCA/Distortion stage, and what one gets out is a horrible mix of really nasty harmonics under most settings of the waveform and brightness controls.
The effect of the VCA Distortion is entirely amplitude dependent, and amplitude is greatest at the low frequencies, so the distortion is really most noticable at low frequencies, but is present right up to 1kHz.
----------------------
PROVISIONAL CONCLUSION
----------------------
The experiments undertaken at the start of all this may well be as good as any simple solution can get.
Reducing oscillator coupling has an effect on linearity at the bass end, so any change to the coupling could have big consequences.
To clean up the sound, the VCA circuit needs to be changed and the distortion components and controls removed - this can be done without needing a piggyback board.
As I see it now, one could reasonably easily make the EW tone a lot more sine-like just by modifying your board.. But you would lose all tone control and could never get a more harmonically rich sound if you wanted it.. I suspect that most people would be unhappy with this trade-off.
There is a way to overcome all the problems, by taking the oscillator waveforms and re-shaping these, then heterodyning them in a good mixer, and feeding them to a good VCA.. This gives great tone possibilities and everything one could want..
Unfortunately, one ends up with quite a large and complex PCB - I know, because I have made one.. And it would be a waste to put this into an EW.*
I will keep thinking about simpler ways to solve the problems (often, after a few days of stewing in the subconcious, and getting a good sleep, the answer just pops