I mostly compared the behavior with TinkerGraph to determine whether there's a difference, but both implementations behave the same for these traversals.
The description of these traversals as being weird comes from me, but I mostly meant a more complicated version which includes more steps which didn't make much sense to me. The traversals are probably generated by cypher-for-gremlin.
But I wasn't sure what the expected behavior is with the limit(0), 2x addV() and the added dedup() in the second traversal which is why my advise was to ask this here.
Is it really intended that dedup() changes the outcome here?