The raw block access list was boring. Besu, Erigon, Ethrex, Nethermind and Reth all returned the same 92,543 RLP bytes for one canonical Glamsterdam block, and those bytes hashed back to the header. Then I asked for decoded JSON: Reth shortened 424 fields that the API defines as exactly 32 bytes.
Every 12 seconds, your execution client gets a new block and has to tell the consensus layer: valid or not. The call is engine_newPayload, and how long it takes determines how fast your validator can attest to the new head.
That window matters. A faster EL client means earlier head votes, better attestation accuracy, and more flexibility in how you run your validator. The timing game, MEV extraction, and late block handling all happen inside this gap.
So which execution client is fastest? The data is messier than the win-rate leaderboards suggest.