Traditional PLM systems built on decades-old software and database technology fundamentally struggle to manage the increasingly complex lifecycle data generated by rising product variability, interdisciplinarity, and heterogeneous IT landscapes. Digital Thread Knowledge Graphs have emerged as a solution, but their database persistence and performance remain open research questions. This study benchmarks three paradigms: a disk-based native graph database, an in-memory native graph database, and an in-memory relational SQL database with a graph extension. The evaluation examines filtering performance, graph traversal algorithms, scalability with increasing graph size, and the impact of connectivity (total average degree) in Digital Thread applications. Results show in-memory native graph databases excel for highly connected datasets and recursive queries, while SQL-based graph extensions are viable for small graphs, existing SQL infrastructures, or non-recursive queries. However, SQL-based solutions show exponential query time growth as graph size increases. These insights guide database selection for scalable, high-performance PLM applications.