This is hard to tell by a single query and schema. A real-world replication would be better.
See, in https://docs.dgraph.io/ we have a simple cluster configuration. And we can run all of the queries in the docs with millions of nquads. If you find a similar query in the documentation with the same problem. It would be easy for us to keep up.
What may be happening to you are numerous factors. I would recommend that you use Jaeger. To trace the source of the problem. Although a bit complicated, it is the best tool for finding bottlenecks and thus helping to better understand this problem.
Donwload it https://www.jaegertracing.io/ and run:
./dgraph zero --jaeger.collector=http://localhost:14268 --expose_trace
./dgraph alpha --lru_mb 8000 --jaeger.collector=http://localhost:14268 --expose_trace