Mixed-phase clouds over the Southern Ocean profoundly influence Earth's radiative balance, yet climate models persistently exhibit a positive surface shortwave radiation bias driven by over-glaciation in mixed-phase clouds. Addressing this bias requires accurate representation of ice-nucleating particles (INPs). This study presents the first online implementation of aerosol-aware marine INP parametrizations derived interactively from sea spray and marine organic aerosols within the high-resolution Unified Model. Using CAPRICORN-2 shipborne observations, we demonstrate that the default INP scheme overestimates INP concentrations by up to four orders of magnitude, causing a severely underestimated liquid water path. In contrast, empirical Antarctic and deterministic marine INP schemes reproduce the low INP concentrations typical of the pristine Southern Ocean, improving cloud and radiative properties. Crucially, microphysical and radiative responses to these INP reductions are strongly regime-dependent. In deep, pre-frontal mixed-phase clouds, suppressing INP concentrations effectively inhibits cloud glaciation, significantly enhancing supercooled liquid water and largely mitigating the surface shortwave radiation bias. However, in shallow, post-frontal stratocumulus clouds, altering INP parametrization yields negligible improvements. Radiosonde evaluations reveal this insensitivity is driven by several model deficits, including systematically smoothed boundary layer inversions leading to excessive dry air entrainment, exacerbated by underestimated cyclonic moisture transport. Consequently, these shallow clouds are thermodynamically starved of water vapor, rendering microphysical INP adjustments ineffective. Ultimately, fully resolving Southern Ocean cloud-radiation biases requires synergistic advancements in representing aerosols, boundary layer physics and large-scale meteorological forcings.