Shame there are no time divisions on the CRO. I think it is just capacitive pick up of the PWM from the other phases. What is the rating of the motor and how long are the leads to it? What I am thinking is that as one half of an output stage goes on, the current rises exponentially until that half output goes off, the back EMF then rises until that half output comes on again where the current rises (or falls) to a different level, that half output then goes off and more back EMF might be generated. The back EMF = L X dI/dT, as L stays the same, the back EMF should then follow the level of the current which it does not.?? If it is back EMF, it will come from a high impedance source, try putting a resistor across each phase, say 10 K, should be enough to severely damp out any back EMF induced oscillations.
Frank