What happens if the mic modeling is used without proper level calibration?

The modeling algorithms are designed to be robust to a mismatch in gain between channels. Although we recommend a 0.1dB gain matching for optimal performance, 1dB or so of mismatch usually works plenty well. A difference, if audible, tends to be most noticeable in the off-axis response. For example, with cardioid the rejection of sound coming from directly behind the microphone might not be quite as good if the gain is mismatched slightly.