I think it is. It looks like one node in an internal cluster could be eating messages…
FYI, as indicated on Dataloss in the backend?, what we did to “solve” the issue is to add a Data Storage integration on the TTN Application and use the REST API on that storage instead of the MQTT interface.