The accelerometer is a 3 axial one, it would be kinda like 6 unidirectional mics, with each pair pointing in opposite directions in axis ortogonal to the other axis. Not all sounds would be converted into motion of the accel so even if the sensitiivity and sampling rate were big enough things would sound kinda different somtimes (i imagine lower frequencies would be captured more loudly than higher frequencies), and there is also the question of the sampling rate, only sounds up to the frequency half the sampling rate can be picked.

Due to the geometry and materials used on the N900 certain frequencies would be absorbed or amplified differently depending on the direction to the source relative to the N900 (combining the oscilations from all axes, or perhaps even from just a single axis and analyzing the frequencies envelope might provide an increase in the direction detection capabilities, kinda like how holophonics trick the brain into hearing sounds from different directions using just two "microfones")