Clusters are important drivers of regional economic growth. Although their benefits are well recognized, research into their evolution is still ongoing. Most real-world clusters seem to have emerged spontaneously without deliberate policy interventions, each cluster having its own evolutionary path. Since there is a significant gap in our understanding of the forces driving their evolution, this study uses a quantitative approach to investigate the role of inventor collaboration networks in it. Inventor collaboration networks for 30 top-performing American metropolitan clusters were constructed on the basis of patent co-authorship data. The selected clusters operate in hi-tech fields: information technology, communications equipment and the biopharmaceutical industry. Starting from a widely accepted hypothesis that the ‘small-world’ structure is an optimal one for knowledge spillovers and promotes innovation effectively, the authors statistically tested the impact of ‘small-world’ network properties on cluster innovation performance proxied by patent output. The results suggest that the effect of the small-world structure is not as significant as theorists hypothesized, not all clusters benefit from the presence of inventor collaboration networks, and cluster performance can be affected by policy interventions. Our analyses also suggest that cluster typology moderates the impact of inventor network properties on cluster innovation performance.