Question about interesting ingestion benchmarking results vs an older ElasticSearch

Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):
OS 2.4.1
ES 5.6

Describe the issue:
It’s not really a problem but I was wondering whether these benchmark results were reliable:

OS 2.4.1:

$ opensearch-benchmark execute_test --pipeline=benchmark-only --workload=geonames --target-hosts=https://my-os-storage --test-procedure=append-no-conflicts-index-only

| Min Throughput | index-append | 6792.65 | docs/s |
| Mean Throughput | index-append | 7445.48 | docs/s |
| Median Throughput | index-append | 7619.81 | docs/s |
| Max Throughput | index-append | 7799.82 | docs/s |

ES 5.6.1:

$ esrally race --pipeline=benchmark-only --track=geonames --target-hosts=https://my-es6-storage --challenge=append-no-conflicts-index-only

| Min Throughput | index-append | 5564.77 | docs/s |
| Mean Throughput | index-append | 6233.43 | docs/s |
| Median Throughput | index-append | 6294.23 | docs/s |
| Max Throughput | index-append | 6393.03 | docs/s |

Configuration:
Kubernetes, both OpenSearch and ElasticSearch deployed inside statefulsets.
8GiB RAM, 4GiB Java heap, 2 cpu cores

Question is, OpenSearch looks about 20% faster with the same workload and resources, is that a pretty reliable ball park figure for relative performance? I realise ES 6 is very old now but it did seem quite a marked improvement.

One thought I had is, maybe compression is on by default now?

Any insights or advice would be welcome. Thanks!