Oncolytic viruses (OVs) are emerging as a promising cancer treatment, as they selectively replicate inside of and lyse tumor cells. The efficacy of these viruses is further enhanced by engineering the OVs to release molecules that stimulate and attract tumor-targeted cytotoxic T cells. The antitumor activity of these enhanced OVs has been shown to be increased when used in combination with other treatments, such as dendritic cell (DC) injections. We have designed a set of mathematical models that are hierarchically fit to experimental data on melanoma-bearing mice treated with various types of non-enhanced and enhanced OVs, both with and without DC vaccines. As the best-fit parameters are determined by fitting to the average tumor response to a given treatment protocol, we also determine estimates of the population variation from the best-fit values by bootstrapping the data. For all bootstrap samples containing parameters within the 95% credible interval, we searched for optimal protocols for administering OV and DC injections in different regions of treatment space. We identified two regions of treatment space (dose of OV and DC injections) for which treatment response is robust, meaning the optimal protocol (as well as other aspects of treatment response) is preserved across bootstrap samples. We also identified the bifurcation region in treatment space for which the antitumor response is sensitive to the bootstrap sample under consideration. Interestingly, this bifurcation point in treatment space closely corresponds to the doses used in the experiments of our collaborators.