It seems you simply didn't catch my point.
I think I did, but disagree with your assumptions stated below.
I claim that a 3 opamp instrumentation amplifier with no gain in the input (no cross-coupling) has almost no CMRR improvement over the one opamp type.
That is true with perfect source balance, since there is no difference in gain between common mode and differential signals.
However, as Whitlock and others have written in detail before, the output impedance of the driving stage and the input impedance of the receiver form a balanced bridge, so the sensitivity to source impedance imbalance is proportional to the input impedance (common mode input impedance, if the circuit is one that can have common mode and differential mode impedances set independently).
Real sources never have 0 Ohm output impedance, and never have 0 Ohm mismatch between the hot and cold driving impedances, so with real world sources, there will be a difference in how sensitive an instrumentation amp and single diff-amp configuration are to common mode signals if the instrumentation amp takes advantage of the ability to have greatly increased common mode impedance as I have suggested.
It is almost equally sensitive to resistor matching errors, which is usually dominant over source impedance unbalance and stray capacitance.
This may be the source of our disagreement. You can get 0.1% resistors at a reasonable price for DIY or boutique devices. If you go to 0.01% the prices get exorbitant, but I like to use the TI INA137 or ThatCorp 1246 devices, which are only a couple of dollars for op-amp with trimmed resistors in the package, and those have typically 0.005% resistor matching. There is never a situation where the combination of source impedance match and stray capacitance balance is better than 0.01%.
If you use those devices directly as input receivers, the very close resistor matching is lost due to source mismatch (of course the matching on the device is not lost, but the important value is the combination of source impedance and receiver impedance, and that will never be matched to 0.005%).
By preceding the integrated differential device with a pair of buffers with high common mode impedance you can better take advantage of the trimmed resistors. If you do not want the differential input impedance to be so high you can put a 10k or 20k resistor across the hot and cold lines to reduce the differential impedance without affecting the common mode impedance. That will also have some small benefit in reducing common-mode to differential conversion.