Sounds a bit unrealistic. Output voltage changes under static load changes can be in the hundreds of millivolts, standardly, under fast load changes even volts! Also, 5mV corresponds to a deviation of only 5mV x 100% / 60V = 0.008%! How to achieve this with real world's components?

The tracking of bipolar voltages puts another interesting question: How to proceed, if one of the voltages breaks down due to the invoke of current limiting? Shall the other voltage follow or stay unaffected?
With standard tracking schemes one of the both voltages is fabricated by the help of the other. But, this would mean an unsymmetrical performance of tracking: Assume for instance, that the positive voltage is derived from the negative. If the negative one now breaks down due to the invoke of current limiting, then the positive voltage would follow. But if the current limiting would be invoked at the positive voltage, then the negative voltage would remain unaffected. Such an unsymmetrical behaviour might not be the one that is wanted or expected, at least from my point-of-view.

I would proceed differently: I would fabricate two identical and very precise supply voltages and link them together at the output to get a bipolar supply voltage. Precision of voltages could be enhanced by using components showing a tolerance of better than +-0.01% and by using of DACs providing a precision of better than 12bit. The voltage of each supply could then be fabricated showing an error of less than about +-10...20mV.
And the tracking is provided by transmitting suited digital data via an opto-isolated serial data link: Two 12bit words, one for the output voltage and one for the current limiting.

Kai