Hi Adrian,
Sorry, but I don't think there is an easy way to predict the accuracy of such a system without an extensive analysis that takes into consideration all the mechanical dynamics, masses, compliances, cutting forces, friction, resonances, etc.
With a backlash of 50 microsteps there is basically no way to guarantee errors to be less than 50 microsteps. This is because any sudden force could instantly move the axis 50 microsteps. The closed loop feedback would then attempt to correct the error. How fast the error could be corrected will depend on the system dynamics.
So in cases where loads and cutting forces occur gradually and the system is low mass and stiff then backlash effects may be reduced somewhat, but otherwise possibly not significantly reduced at all.
As a rule of thumb to achieve a certain precision/accuracy of X it is best to have a measurement and movement resolution of at least 10X better. So attempting an accuracy of 10 with measurement of 5 is unlikely to work. The main need for higher measurement resolution it to know velocity. In that case a motion of 5 microsteps is required to have any information on whether the axis is moving. And then there will be virtually no information on its velocity.
HTH