Dear John M,
unfortunately John Reekies' workaround does not work in our case which is:
MBPro OS X 10.8.2
Focusrite Saffire 40 PRO, 44.1kHz
Jave updated to latest version
REW 5.0, 44.1kHz
Since we could not get FireWire to work for the input (output worked fine), we configured only the output via FireWire (then to Sub, Crossover, Speaker, Room, Mic) and came back from the Saffire via its Analog Out to the MBPro LineIn, which seemed to work in the beginning.
I started with 2 sweeps per measurement and got caught by a 30dB dip around 11k in the FR.
To analyse, we spared out the Speakers and the Mic, connected the two cables directly to each other ('wired loop') and found in a mostly flat response the mentioned dip.
After quite a while and mostly by chance I found that reducing the sweeps to 1 per measurement gives me a flat result. See the 1sweep vs. 2 sweeps in the next image
Diggin deeper I wondered, what happened, if I'd increase #of sweeps, and here's what I got for 4 sweeps
and here for 8 sweeps.
I only guessed, that, since these results are very reproduceable, it's got nothin to do with WiFi or Laptop CPU overload or sg. - nevertheless I checked, turned everything off, dips are still there.
Then I came across the above mentioned input workaround. We did it according to the descriptions, but what we got was, instead of getting better, that signal generation, output and input "reacting" extremely delayed. With the meters on the screen, after hitting check levels, or measure, or calibrate we could wait and then see with more than a sec delay the REW pre OUT meter showing the signal, then, approx another sec later the focusrite input meter, and then, slowly rising, the REW pref In meter - showing the signals.
We got error messages, that the IR is not at the expected position, so we tried a loopback as a t0 reference, which gave us a flat result till 3kHz, but then jumped down into noise (sorry - no pic from that).
After restarts, of Saffire, MBPro, REW etc and no improvment, we were forced to go back to our initial solution and uninstall the input workaround (Soundflower).
So we are still stuck with the limitation to only being able to do 1 sweep, which I'd like to overcome, to improve S/N, apart from increasing the sweep length, since it is a bit noisy around where we have to measure...
To me, it looks like an issue related to either windowing or aliasing or similar, but I did not manage to put my finger on it...
Any ideas regarding our dip-pattern, John M maybe?
Many thanks for your thoughts on that!
Markus