Under a single-copy Helstrom distinguishability budget, when do noncommuting quantum states offer a conditional work advantage over commuting ones? For a four-level Hamiltonian H = Δ diag(0, a, b, 1), we optimize both the preparation priors and conditional states, obtaining exact commuting and unrestricted work frontiers. The two frontiers coincide in three or fewer dimensions; in four dimensions a strict advantage exists for some a, q exactly when b > 9/11. The mechanism is a constraint on the conditional eigenvalue weight attainable by commuting ensembles. In the strict region every optimum belongs to a single unitary orbit with rank-two conditional states and equal priors. A coupled-qubit example illustrates the effect without entangled conditional preparations. The bounds concern ideal feedback-assisted work capacity, rather than net cyclic output.