Linked data can be represented as graphs, and core graph-based tasks are required not only for consuming linked data, but also for mining associations and patterns among data and links. To facilitate these tasks, efficient algorithms have been defined; additionally, graph database engines that manage, store and query large graphs have been implemented. Nevertheless, there is no clear understanding of how graph-based tasks may behave on these systems. In this paper we evaluate general purpose graph database engines and one state-of-the-art RDF engine. Our experimental results reveal properties of these systems and the characteristics of the graph-based tasks that may affect their performance. These results can be considered as a further step for solving the problem of choosing between graph databases to consume and mine linked data.