The return 208 from the converter has to be in-phase with the 480. It also has to be the correct voltage at whatever the half tap is. If not for either case will result in neg seq current (potentially excessive) that will over heat the xfmr and lead to failure. By using a different xfmr to convert the 208 back to 480 allows you to use the impedance of the xfmr to your advantage and maintain isolation. The easiest way to test the OPs and Besoeker's design is to connect the system without the return 208 and spin it up. Then, measure the potential difference between the return 208 and the half tap on the source xfmr. Any difference in voltage will result in neg seq current flow.