Due to the distributed nature of Linked Data, many resources are referred to by more than one URI. This phenomenon, known as co-reference, increases the probability of leaving out implicit semantically related results when querying Linked Data. The probability of co-reference increases further when considering distributed SPARQL queries over a larger set of distributed datasets. Addressing co-reference in Linked Data queries, on one hand, increases complexity of query processing. On the other hand, it requires changes in how statistics of datasets are taken into consideration. We investigate these two challenges of addressing co-reference in distributed SPARQL queries, and propose two methods to improve query efficiency: 1) a model named Virtual Graph, that trans- forms a query with co-reference into a normal query with pre-existing bindings; 2) an algorithm named, that intensively exploits parallelism, and dynamically optimises queries using runtime statistics. We deploy both methods in an distributed engine called LHD-d. To evaluate LHD-d, we investigate the distribution of co-reference in the real world, based on which we simulate an experimental RDF network. In this environment we demonstrate the advantages of LHD-d for distributed SPARQL queries in environments with co-reference