Consider the σ -finite measure-valued diffusion corresponding to the evolution equation u t = Lu + β(x)u − f (x, u), whereand n is a smooth kernel satisfying an integrability condition. We assume that β, α ∈ C η (R d ) with η ∈ (0, 1], and α > 0. Under appropriate spectral theoretical assumptions we prove the existence of the random measure(with respect to the vague topology), where λ c is the generalized principal eigenvalue of L + β on R d and it is assumed to be finite and positive, completing a result of Pinsky on the expectation of the rescaled process. Moreover, we prove that this limiting random measure is a nonnegative nondegenerate random multiple of a deterministic measure related to the operator L + β.When β is bounded from above, X is finite measure-valued. In this case, under an additional assumption on L + β, we can actually prove the existence of the previous limit with respect to the weak topology.As a particular case, we show that if L corresponds to a positive recurrent diffusion Y and β is a positive constant, then lim t ↑∞ e −βt X t (dx) exists and equals a nonnegative nondegenerate random multiple of the invariant measure for Y .Taking L = 1 2 on R and replacing β by δ 0 (super-Brownian motion with a single point source), we prove a similar result with λ c replaced by 1 2 and with the deterministic measure e −|x| dx, giving an answer in the affirmative to a problem proposed by Engländer and Fleischmann [Stochastic Process. Appl. 88 (2000) 37-58].The proofs are based upon two new results on invariant curves of strongly continuous nonlinear semigroups.