I googled permanent magnet DC starter motor running backwards, and ran across this post made by some guy named Ebb's Speed Shop. I did not post this, but I did copy / paste it here.
A little side note: A lot of DC motors are optimized to run one direction, and are built based on the RPM the motor is designed for. DC motors have "timing" believe it or not. The timing is related to the position of the permanent magnets to the armature, and of the brushes to the commutator. Ideally, you want the magnetic flux in the armature to peak right when the poles of the armature are in the midst of the strongest part of the permanent magnet's flux field. So there is a little "advance" built into this timing, since it does take a short time for the flux to build in the armature. Now if you take this same motor and reverse the polarity, it will run backwards, but it will not produce the torque nor the speed that it does when running in the "proper" direction.
This is the thread:
www.jalopyjournal.com/forum/threads/can-...ors-polarity.259628/
And yet another copy / paste from the Rad One: (note, I did not edit all the typo's, but chrome did highlight them all for me.)
Most DC motors I've messed with will turn the other way when reversing the polarity. But to get the same torque or rpm in the reverse direction you have to be able to adjust the timming. Timming is the relationship between where the brushes come in contact with the commutator segments to where the field coil reacts with the armuture windings. The starter bellhousing were the brushs sit must be able to rotate to do this and all of them are fixed with index notches and bolts.
Link to the thread:
forum.miata.net/vb/archive/index.php/t-104550.html
It sounds to me as if the magnets are not in their exact positions, because everything else is set with notches and pins to locate them in relation to the armature and magnets.