> Since it is spent it does not bloat the mempool.
 
This is not the case. If you post some 100 kB TapScript, with some Ordinal, then it of course bloats mempools, because then other users could post 100 kB less, when it comes to regular payments. If you have Ordinals in the current form, then they take place of regular payments. Which means, you can include some payment, or some data. You cannot include both, because you can produce 4 MB block per 10 minutes. It is always a choice: confirm this payment, or confirm that data.
 
> Regardless of OP_RETURN the data will be on chain. What am I missing?
 
If you have regular OP_RETURN, the data is pushed on-chain. However, if your OP_RETURN is part of your TapScript, then you cannot provide any valid input to satisfy that kind of TapScript, so it cannot be pushed on-chain. Which means, you have to use another TapScript branch, without OP_RETURN, or simply spend by key. Note that regular OP_RETURNs are placed in transaction outputs, but in that case, TapScript is revealed in transaction input (and to be more specific, in the witness part), which prevents from posting a commitment on-chain, if it contains OP_RETURN at the beginning of the TapScript.
 
> If there was no need for people in this list to discuss it before I don't see why a BIP is needed now.
 
It is needed, but for a different reason. There should be a BIP, but not to promote Ordinals, but to handle them properly, and to provide regular users a way, to compete with the currently posted Ordinals, in this fee-based competition. Which means, regular users should have a way, to turn their Ordinals into proper commitments. They should be hidden behind some R-value, instead of being posted as a TapScript, and fully revealed in the witness. That would make it smaller, cheaper, and will provide more room for more regular payments, while preserving the strong cryptographical proof, that a given data is connected with a given transaction input.
 
Also, it would allow them to be uncensorable, because then users could decide to hide any Ordinal behind their signature, in any address type (it works even on P2PK), and then reveal it later, but not on-chain, to not take a room of other regular payments. In general, transactions should always contain payments, and Ordinals could be attached as a feature, and not the other way around, when the actual payment is just a feature to be discarded, and the posted data is what people care about. Bitcoin is a payment system first, and not a P2P cloud storage, so it should work as "always a payment, and optionally also an Ordinal", and not "just a data push, and unfortunately a payment, because the protocol forced us to include some satoshis".