Why is this `pgbench` benchmark faster on postgresql 11 than on postgresql 14?

To check postgresql performance, I ran the following benchmark with varying postgresql versions and filesystems:

pgbench --initialize --scale=50 bench
# Run for 3 minutes
postgres pgbench --client=10 --jobs=2 --time=180 bench

#### System environment - postgresql settings: shared_buffers = 8 GiB (of 32 GiB total system RAM) - 2x 7200 RPM HDDs #### Results | Filesystem | TPS (postgresql 11.16) | TPS (postgresql 14.4) | | ---------- | ------------------- | --- | | zfs compression=lz4 | 497.21 | 288.93 | | zfs compression=lz4 recordsize=8K | 488.11 | 430.83 | | ext4 | 492.01 | 480.66 | | btrfs mirror compression=zstd | 193.65 | 187.4 | Overall, pg 11 is faster than pg 14, especially on ZFS with the default 128K recordsize. What's the explanation for this result? Did pgbench change between versions? Or did postgres increase data robustness/crash resistance at the cost of performance?