Marginal ROAS asks: when spend went from one level to the next, what did the additional cohorts cost, and what are they returning at equivalent maturity? It is the right metric for budget allocation, because returns usually diminish as campaigns reach less responsive audiences.
It is not obtained just by slicing blended ROAS by source. You need to isolate the incremental spend and the incremental users it produced, ideally by comparing cohorts at the same age before and after the step change.