For the record, I came up with a couple of ways to achieve my goals (measuring frequency response, phase response, and group delay of a digital filter). First, I managed to perform the analysis in MATLAB. But it's too much geeky and DIY (although, very much educating).
Then, in order to check the results obtained from MATLAB, I've managed to "fool" REW and feed to it a pre-recorded processed response for a measurement sweep. For that, I used Rogue Amoeba Loopback in "pass-thru" mode (on Mac). I set up REW's input from the virtual device, and then I was simply playing the response .wav file in VLC into the loopback device while REW was doing a measurement. Worked like a charm!
And the third approach (to triple-check) was to set up the filter under investigation in "real-time" processing mode on one computer, and route REW's measurement sweep via a real soundcard connected criss-cross with the sound card on the computer running the filter. This produced the noisiest results, but in general they also agreed with the previous two.
One shortcoming I've encountered with REW is that is generally expects delays in milliseconds for GD (I guess, it's normal for acoustic systems), while my filter produces microsecond delays. Although REW is showing pretty similar numbers, they are not as precise as those I was able to calculate with MATLAB.
But it's great that it all worked and now I actually have understanding of how REW does its graphs!
One thing I'm still amazed about REW is how fast it can perform 1/N-octave smoothing. :T I've found one packet for MATLAB that does that, but it's su-u-uper slow, I'm not sure why.
John, is it possible for you to give a hint how REW is able to perform smoothing so quickly?