For comparability of results it would be useful to have all sub-runs be grouped into a single larger job so that all scheduled nodes are the same for the sub-runs.
It would be another dependency, but you could also consider integrating with the bolt tool for run script generation: see https://github.com/EPCCed/bolt