The calibration of complex models of biological systems requires numerical simulation and optimization procedures to infer undetermined parameters and fit measured data. The optimization step typically employs heuristic global optimization algorithms, but due to measurement noise and the many degrees of freedom, it is not guaranteed that the identified single optimum is also the most meaningful parameter set. Multimodal optimization allows for identifying multiple optima in parallel. We consider high-dimensional benchmark functions and a realistic metabolic network model from systems biology to compare evolutionary and swarm-based multimodal methods. We show that an extended swarm based niching algorithm is able to find a considerable set of solutions in parallel, which have significantly more explanatory power. As an outline of the information gain, the variations in the set of high-quality solutions are contrasted to a state-of-the-art global sensitivity analysis.