Currently the query service is balanced between two node.
Those two nodes currently return slightly different data in slightly different cases.
An example of this is the total number of triples stored.
This can currently be seen at https://grafana.wikimedia.org/dashboard/db/wikidata-query-service
Having the server name in the response header would mean we could track this individually
It would also make debugging easier in the long run