Differential grouping (DG) is a pregrouping method in some unconstrained cooperative coevolutionary algorithms where the grouping scheme is predetermined, while random grouping (RG) is a method where the grouping schemes are dynamically changed according to some criteria. As the first nature-inspired method for large-scale constrained optimization, the ε-constrained cooperative coevolutionary particle swarm optimization εCCPSO) algorithmic framework has been recently proposed and studied with different RG techniques, e.g., εCCPSOw2. This paper aims to compare the effectiveness of DG and RG methods applied on the εCCPSO framework, using εCCPSOw2 and two proposed DG implementations, i.e., εCCPSOdg and εCCPSOdg2, for large-scale constrained problems. The results show that, while εCCPSOw2 possesses the fastest convergence rate on the fitness values of most problems, εCCPSOdg2 has better constraint violations minimization ability when solving some problems with nonseparable equality constraints. In addition, an improvement of the DG method is proposed, increasing the speed to approximately twice the original.