1) I too would like to see some of the data differences delineated, along with some rows that are identical. Please share any patterns you note in the results.

2) Is is possible that your joins are not 100% sufficient?

3) It is quite possible that the fix did NOT actually make it into the higher releases as ON by default. Did you try TF 4134 on both builds you are testing on?

4) Got the actual execution plans by chance?