But I want to test it into a differential input ( eg 'standard 4 resistor diff amp ). I just can't get it straight in my head (or on the screen) how I should configure this wrt injecting CM noise. Any thoughts ?

Just model a virtual balanced cable which could be as simple as connecting pin 2's and pin 3's together.

But I think CM performance is going to depend largely on how closely matched the two impedances are (actually 4 impedances - 2 out and 2 in).

To understand that, I would do an impedance plot. An impedance plot is like the usual AC analysis with the usual signal generator, except instead of plotting V over frequency you plot V/I over frequency (recall that Ohms law says that R = V/I and so in AC analysis R means impedance). Specifically, add a load resistor to ground on each pin and then do AC analysis, Run and click on the output pin to plot V which will show a data series label at the top like "V(n005)". Right click on that and edit the formula to read "V(n005)/I(R7)" where R7 is your load resistor to ground [1] and click Ok. Now right click on the scale on the left and select "Manual Limits". Change the drop down from "Decibel" to "Linear". Now the V/I axis should show impedance! Now click on the other output pin and edit it's formula. Now you can see the difference in impedance and decide how significant it is.

[1] If the resistor isn't grounded, you would have to use the difference in voltage across the resistor which changes the formula a little be equally useful. Actually the officially proper way to do this is to replace the signal generator with a load resistor (like 150R for a mic input for example) and attach an AC current source to the output(s) and then do like above but use a formula like "V(n005)/I(I1)". If you search "ltspice plot impedance" this is usually how it's described.