In that thread, you mentioned that the firmware is supposed to accelerate/decelerate at the specified rate - have you verified that it does? Your test case of a full axis move isn't exactly real-world (plenty of t for the dx to change over) - what happens when decelerating around a 2mm 45 degree notch for instance - does X veloctiy decrease so Y can, or does the x velocity stay constant, and the Y velocity instantaneously invert? On my machine the banging suggests the latter case (I presumed it was the head snarling on pridges on the print layed down in previous levels, but running with no filament produces similar results)
Ray