If we have the option to rotate the camera in Z axis, we can manually fine tune the calibration scene,
and have the option to work with kinect at 90º angle to record on very small places (full body capture starts about 1,7m from kinect)

I made some tests with the xml file, it works but of course it's not viable.

It sounds a bit odd relative to calibration :) Automatic detection of camera positions should always produce more accurate results than manual adjustments when given an appropriate input data.

Support for vertical placement of depth sensor will require changes in the ground detection algorythm. And, possibly this will involve manual specifying the area on the picture where the ground is located. I think we can consider this only for the next major version.
Anyway, thanks for sharing!