I was skeptical about my  previous mode matching model , which indeed was not right. 
I didn't have the correct SR2-SR3 and SR2-SRM3 distances ( E1200616 ) calculated according to the H1 SR3 and SR2 ROCs. 

Based on lessons learned from L1, we expect the matching to the OMC X/Y in single bounce to be determined by the different ITMs ROC, while the overall scale is set by the combination of PR3(SR3) ROC + PR2-PR3(SR2-SR3) distance. 

The "theoretical numbers" for the PR2-PR3(SR2-SR3) distance calculated on the base of the measured PR3(SR3) ROCs should give us a mode matching close to optimal.

Indeed, the theoretical numbers for the mode matching are closer to perfect mode matching, as it should be: 

single bounce X = 96%
single bounce Y = 99.5%

Because of the uncertainty in the PR3 and SR3 ROCs, and in the PR2-PR3(SR2-SR3) distance, the overall matching can be somewhat worse, and the X/Y asymmetry should be amplified by the SR3 ROC/ SR3-SR2 length mismatch. 

For example, a mismatch SR3 ROC / SR2-SR3 length equivalent to 1 cm error in SR2-SR3 length would give us:

single bounce X = 90%
single bounce Y = 96%

A mismatch SR3 ROC / SR2-SR3 length equivalent to 2 cm error in SR2-SR3 length would give us:

single bounce X = 82%
single bounce Y = 90%

And so on. 

However, based on  Koji's calculations, we see pretty much the same mode mismatch X/Y (12%-13%). 

So, the message is that we can't really predict the overall matching to the OMC, given the uncertainty in the recycling cavity parameters. However, we should be able to see an asymmetry X/Y due to the different ITMs ROCs. The measured OMC matching numbers are not unreasonable (12%-13%), but they don't show the X/Y asymmetry that we expected. I am not sure why, but probably we don't care now. We might at some point want to repeat an OMC X/Y scan with dither on to have more precise measurements.