This study examines the impact of collaborating patterns on the R&D performance of public research institutions (PRIs) in Korea's science and engineering fields. For the construction of R&D collaborating networks based on the co-authorship data of 127 institutions in Scopus, this paper proposes four types of collaborations by categorizing network analyses into two dimensions: structural positions (density, efficiency, and betweeness centrality) and the relational characteristics of individual nodes (eigenvector and closeness centralities). To explore the research performance by collaboration type, we employ a data envelopment analysis window analysis of a panel of 23 PRIs over a 10-year period. Comparing the R&D productivities of each group, we find that the PRIs of higher productivity adhere to a cohesive networking strategy, retaining intensive relations with their existing partners. The empirical results suggest that excessively cohesive alliances might end up in 'lock-in' relations, hindering the exploitation of new opportunities for innovation. These findings are implicit in relation to the Korean Government's R&D policies on collaborating strategies to produce sustained research results with the advent of the convergence research era.