This study examines whether candidate-level gain–loss information improves search for cost-aware portfolio rebalancing. We propose Risk-Sensitive Foraging Artificial Bee Colony (RSF-ABC) for long-only portfolios with exact cardinality and active-weight bounds. A bounded state derived from the Omega objective controls the probability of heavy-tailed exploration versus conservative best-guided search. A state-independent mixture of the same operators isolates the value of state conditioning. Monthly evaluation on U.S. industry portfolios uses 504-day formation windows and proportional transaction costs. Mean net Omega is 1.1635 for RSF-ABC, 1.1284 for Standard ABC, 1.1342 for Fixed-Mix-ABC, 1.1410 for particle swarm optimization, and 1.1450 for differential evolution. Monthly turnover is 0.4780 versus 0.5570 for Standard ABC. The proposed method leads in five of seven boundary conditions; deterministic alternatives lead in two. At 49 assets, its reported median runtime is 1.11 s versus 1.08 s for Standard ABC. These results support the evaluated state–operator coupling.