ArangoDB crashes with two hop graph query.

Image three vertex collections VA, VB and VC. VA is connected to VB and VC. We have an instance of VB and want all VC that are connected to it via edges and VA instances.
All of them are definded in a graph to access them with the Arango graph API. I used the following AQL statement to query the graph. First I get all VAs that are connected to the VB instance and then all VCs that are connected to the VAs.
FOR va IN GRAPH_NEIGHBORS("Graph", "VB/Instance", {direction: "inbound", edgeCollectionRestriction: "eAB"})
FOR vc in GRAPH_NEIGHBORS("Graph", va._id, {direction: "outbound", edgeCollectionRestriction: "eBA"})
RETURN vc
The result is that after a few minutes of computation the ArangoDB crashes without any useful information in the log files.
It seems very inefficient to model edges as documents and not links between documents, because in such a query the whole edge collection has be run through multiple times to find the correct links. I assume that two hops is just to much for the database to handle. Or is there any potential to improve the query and not crash the database?
GRAPH_NEIGHBORS. Can you try if the following query is any better:FOR va IN NEIGHBORS(VB, eAB, 'VB/Instance', 'inbound') FOR vc IN NEIGHBORS(VB, eBA, va.vertex._id, 'outbound') RETURN vc.vertex? Apart from that, there have been some changes in 2.6 to makeGRAPH_NEIGHBORSandNEIGHBORSrun a lot faster than in previous versions, which may also fix this issue. - stj