2 data prepper node work slowly than 1

Data Prepper
1

Versions
opensearch-2.18.0-1.x86_64
opensearch-data-prepper-jdk-2.10.2-linux-x64

Describe the issue:
2 Ingest nodes with data prepper, read a Kafka topic of simple log data (dummy log with a @timestamp).

I run only ingest_1, all is ok, I’m limited a 30K log/sec.

When I add the second ingest node, ingest_2, performance drop down.

I need 2 ingest for disponibility requirement.

Why adding a second ingest prepper node drop down the speed?

How could I fix it?

Configuration:
kafka: 10.10.10.100:9092
cluster: 6 data, 2 master & 2 ingest nodes.
2 ingest nodes:
ingest_1 : 4 vCPU 10Go RAM 10.10.10.1
ingest_2: 2 vCPU 10Go RAM 10.10.10.2

Relevant configuration:
Same configuration for ingest_1 & ingest_2, except workers.

pipelines.yaml

---
kafka-pipeline:
  workers: 8 # 4 for ingest_2
  delay: "50"
  source:
    kafka:
      bootstrap_servers:
        - 10.10.10.100:9092
      topics:
        - name: pipeline-fluentbit
          group_id: opensearch-data-prepper
          auto_offset_reset: "earliest"
          auto_commit: true
      encryption:
        type: "none"
        insecure: true

  processor:
    - parse_json:
        source: message
        delete_source: true
    - date:
        from_time_received: true
        destination: "@timestamp"

  sink:
    - opensearch:
        hosts:
          - "https://10.10.10.1:9200"
          - "https://10.10.10.2:9200"
        index: "pipeline-fluentbit"
        username: "admin"
        password: ""
        insecure: true
        batch_size: 5000
        flush_interval: "50"
---