Stokes polarimeters measure some or all the Stokes parameters of a light beam. To reduce the effect of noise in the measurements made on the final calculated Stokes parameters, a polarimeter can be optimized by minimizing the condition number of the characteristic or instrument matrix. However, this optimization does not guarantee the best stability in the presence of experimental errors in the polarimeter configuration. This is particularly important for polarimeters using liquid-crystal variable retarders, which have we have found have errors in the retardance as a function of position in the aperture of the liquid-crystal cell, and fast axis angles as a function of the applied voltage, so even the best aligned of this type of system will have these errors. We have found that different optimized polarimeter configurations can have very different sensitivity to experimental errors. In this work, we present an analysis of a number of different optimized systems to find the most stable configuration with respect to experimental errors. In particular, we consider the variation of the volume of the solid formed in the Poincaré sphere by the Stokes vector values used in each polarimeter configuration.