PS: I was talking to Florian Walter (the driving force behind MIR's convolution engine). He kindly asks you to send the two simple test-setups you were describing above to , with something like "For Florian" in the subject line. He will look into it as soon as he finds the time. Please include all the information that is necessary to recreate your setup as closely as possible. The main difference will be that he uses PT 10 Native, not HD, though.
Best,
/Dietz - Vienna Symphonic Library