Operations per second drop every 10 minutes

I’m currently testing out Couchbase using a 4 nodes Couchbase cluster on VMs.
Durability level: majority
Replication: 1
Ejection method: value-only

I am running a test with ~20k(12k GET/6k SET/2k EXIST) operations per second. Every 10 minutes a significant drop to 1k operations per second is observed. At the time of drop, the “DCP Replication Items Remaining” increases to 700 to 350:

During the test, we have the following stats:
items: 125,448,930
resident: 4.62%
RAM used/quota: 37.8GiB / 48GiB
disk used: 307GiB

What could be the cause?