Performance gain of the hybrid scheme (hSIDM) developed in this work for taking the angle dependence of DM self-scattering (characteristic for e.g. light mediator models) into account in N-body simulations. We show the CPU run-time t_cpu (executed on a single CPU only) required for advancing the deflection test problem up to physical simulation time t (in units of the typical reference timescale t₁) in the upper panel. The red line corresponds to the naive scheme for which scatterings by all angles are sampled from dσ/dΩ (anisotropic rSIDM). The CPU-time within the hSIDM approach is shown for runs with various values of the critical angle θ_c used to separate effective small- and explicit large-angle regimes. The CPU-time is reduced by almost a factor 100 for θ_c ∼ 0.3, and already significantly smaller for θ_c ∼ 0.1. The lower panel shows the amount of CPU run-time Δt_cpu required to simulate a given physical simulation time interval Δt_sim, which is roughly t-independent. The y-axis is normalised to unity for the anisotropic rSIDM scheme. The CPU run-time decreases strongly for θ_c ≲ 0.1, and saturates for θ_c ∼ 0.3. For even larger critical angles, the numerical cost of the effective small-angle approach becomes relevant as compared to the explicit treatment of large-angle scatterings within hSIDM.