I'm looking for information on how X3DAudio maps 3d sounds to speaker positions.  I realize it's done by setting the pMatrixCoefficients member of X3DAUDIO_DSP_SETTINGS, but when you specify X3DAUDIO_CALCULATE_MATRIX, how is it handled? Is there a perceptually correct method for defining this matrix, or is it decided on more by preference than correctness? 

I can imagine how one might implement such a system; You could choose a fall off model, specify the speed of sound, whether you're using doppler, and so on, and do calculations based on the recommended geometry for a 5.1/7.1 speaker setup.  Is this how X3DAUDIO_CALCULATE_MATRIX works?

While a general understanding of how the 3D audio to speaker output is done is good, I need specifics.  We're attempting to study audition in free field, and we need to know if X3DAudio will meet our requirements.

X3DAUDIO_DSP_SETTINGS receives the results from a call to X3DAudioCalculate.  The application is just responsible for allocating the buffer needed to receive X3DAudio’s output.  X3DAudio is responsible for generating the matrix values dependent on the 3D parameters the application sets in X3DAUDIO_LISTENER, X3DAUDIO_EMITTER etc. 

X3DAudio currently uses very straightforward positioning logic: it distributes the signal to the nearest two speakers in the direction of the emitter as projected onto the listener’s XZ plane.  Elevation cues may be applied via X3Audio’s InnerRadius/InnerRandiusAngle functionality if desired as explained here:
http://msdn.microsoft.com/en-us/library/bb174736(VS.85).aspx