His point is that the robot has split the kinematics to have 3 DOF for position near the base and 3 orientation DOF on the wrist.

This still gives you 6 DOF on the end effector, which is pretty good, but overall the arm design is restricted in its ability to route around obstacles.

This downside has an upside though. Since every cartesian position has exactly one pose for the first 3 DOF, the inverse kinematics are simpler and you do not run into singularities for basic position control when maintaining a constant orientation.