Hi Christopher,
Some feedback:
"OP_RETURN is limited to 40 bytes of data."
It is 80 bytes.
"A future BIP proposing such a layer two protocol will be forthcoming."
So what is this BIP about? Just saying that it would be a nice idea? This BIP should be the one that would go through this L2 suggestion. If one root OP_RETURN substitutes all the rest it should say how that would be done... where would the merkle proofs be stored, what are the trust assumptions that we need to make, etc.
"Objections to this proposal" section
I agree with others re hard-fork, which would be a good thing of course. My main objection with this proposal is that I don't see a proposal. It seems like wishful thinking... if only we could substitute all the OP_RETURNs with one :-)
We have to make sure that a proposal like this (L2, etc.) would make sure that there are incentives that justify the added complexity for the users. Multisig is not the only way data could be stored the wrong way; P2PK, P2PKH, P2SH, P2WPKH, P2WSH can also be used. If the incentives are not good enough people would start using these UTXO-bloat-heavy alternatives.
There are a multitude of L2's (kind-of) that do this 'aggregation' of data hashes using merkle trees. Factom is adding a single merkle root per bitcoin block for the millions upon millions of records (of thousand of users) that they keep in their network. Opentimestamps, tierion, blockstacks and others do a similar thing. I have investigated several of those in the past, for one of my projects, but I ended up using plain old OP_RETURN because the overhead of their (L2-like) solution and trust assumptions where not to my liking; at least for my use case. They were pretty solid/useful for other use cases.
Unless the proposed L2 is flexible/generic enough it would really prohibit this L2 innovation that OP_RETURN allowed (see above).