I share some of the same concerns that this might have the opposite of its intended effect, but I’d like to focus on human behavior during consensus bugs for a minute, while the Nethermind split is still fresh in our collective memory.
During the bug, Nethermind users were faced with an immediate dilemma (whether or not they realized it).
- Switch to a different client, risking participating in a finalizing noncanonical chain if Nethermind happens to be correct.
- Face downtime penalties on the canonical chain if Nethermind happens to be incorrect.
In this case, the safest option is generally (and was) option 2. By correlating attestation penalties, however, you amplify the losses these risk-averse operators incur, and increase the likelihood that a non-canonical supermajority forms in the event of a consensus split.
Fortunately, geth/besu were canonical, and nobody got slashed for switching from nethermind to a geth fallback, but it isn’t a given.