We extend our study of the optimization of large baryon acoustic oscillation (BAO) surveys to return the best constraints on the dark energy, building on Paper I of this series by Parkinson et al. The survey galaxies are assumed to be pre-selected active, star-forming galaxies observed by their line emission with a constant number density across the redshift bin. Star-forming galaxies have a redshift desert in the region 1.6 < z < 2, and so this redshift range was excluded from the analysis. We use the Seo & Eisenstein fitting formula for the accuracies of the BAO measurements, using only the information for the oscillatory part of the power spectrum as distance and expansion rate rulers. We go beyond our earlier analysis by examining the effect of including curvature on the optimal survey configuration and updating the expected ‘prior’ constraints from Planck and the Sloan Digital Sky Survey. We once again find that the optimal survey strategy involves minimizing the exposure time and maximizing the survey area (within the instrumental constraints), and that all time should be spent observing in the low-redshift range (z < 1.6) rather than beyond the redshift desert, z > 2. We find that, when assuming a flat universe, the optimal survey makes measurements in the redshift range 0.1 < z < 0.7, but that including curvature as a nuisance parameter requires us to push the maximum redshift to 1.35, to remove the degeneracy between curvature and evolving dark energy. The inclusion of expected other data sets (such as WiggleZ, the Baryon Oscillation Spectroscopic Survey and a stage III Type Ia supernova survey) removes the necessity of measurements below redshift 0.9, and pushes the maximum redshift up to 1.5. We discuss considerations in determining the best survey strategy in light of uncertainty in the true underlying cosmological model.