Alberto
you a right when you say that a similar process is behind the image rejection and the DSB/SSB noise difference.
The reason is that noise coming from common part of the receiver (not the one generated by I & Q path) that is usually the dominating one, is correlated, so it is canceled out by proper IQ recombination (that, in your case, should be done in a complex digital IQ mixer).
First, did you check that you have image rejection? How many dBs?
Second, can you show briefly a cascade analisys of the receiver?
I've designed (for large volume products) both image rejection mixers and low IF receiver, and, in both cases they worked as expected.
I hope it can help.
Mazz