The size, complexity, heterogeneity, and dynamism of largescale computational grids make autonomic grid services and solutions necessary. In particular, grid schedulers must map applications onto resources whose state (1) influences the effectiveness of scheduling choices, and (2) changes frequently and considerably. A grid resource state information dissemination service must negotiate the inherent tradeoff between covering a large portion of the grid (so that all schedulers can make informed decisions with the largest number of options), and limiting the protocol?s overhead (i.e. the number of packets sent). This paper argues that probabilistic forwarding protocols must adapt to state changes, because static assignments of forwarding probabilities lead to excessive overhead or lower-than-possible query satisfaction rates in some scenarios. We introduce an approach that compares a node?s local utilization and query generation rates to corresponding rates in the node?s vicinity, and in the grid as a whole. These comparisons, in turn, produce a score that is used to adjust forwarding probabilities. We show that even this simple initial adaptive approach can work better than protocols with static forwarding probability assignments.