It now works, and works well.
Thanks Jerry, the different topology between T2 and T4 was the key issue with the unwanted unbalancing. I can't see why, but that is exactly what is happening.
So the successful modification to get the clock #1 mixer to unbalance on CW key down is the following:
1. Re-wire the T4 transformer input and output as per T2. That means as Jerry said to "... cut traces to T4 pins 3,5 and 6.
Swap them so T4 pins 3,5 are in from R47, pin 6 is out to C211."
2. Disconnect (cut trace) from R105 to the common connection of C10, R27 and T2 (3,5).
3. Connect via a wire the disconnected side of R105 to the common connection of T4 (3,5), C211.
I used an audio shielded cable, and passed the wire under the board by drilling two small holes next the two connection points to ensure minimal pick-up of RF.
Pictures of R105 trace cut and wire to T4. (partially obscured by the hot melt glue on the toroids as I use my rig /P and /PM).
The difference in signal strength between key-up and key-down as seen on a control receiver is from way below S0 to S9+20dB or so, giving a dynamic range of over 70dB (indicative value in light of the basic test method).
The output power in CW mode can now be controlled finely by shifting the 45MHz clock slightly along the slope of the Xtal filter. For example I go from 14W max to between 1.5 and 2 watts by shifting the clock #1 frequency by 30Khz on the 20M band. On the 10M band, I need a shift of about 10Khz to reduce the 8 watts out to the same level.
Now to a bit of programming and I have full control of SSB as well as CW power across bands. Great for my built-in ATU.
If someone with a spectrum analyzer could check the implications for the harmonics and spurs that would be a plus. I would expect in CW mode that the harmonic for 80, 40 and 30M bands would reduce, but some spurs to appear since we have the beat of two clocks now. If there is interest I can modify Farhan's code to match that modification (or publish some code snippets).
By the way, before I did the mod above, I also tried to put a trim-pot between R105 / Ground / slider to T4 (3,5) and even when turned all the way to zero (slider to ground) it would still constantly unbalance the clock #1 mixer. So the change in topology as described above is the only way I could get it to work as intended.
73, John (VK2ETA)