Thanks for this (saw this by Richard Gendal Brown).
I agree with your overall thesis but think the issue is more subtle than you’re suggesting. The bitcoin transactional model has a particular power to it, by making the dependencies between transactions localized and explicit. This has immediate value (e.g. trivial rollback) in terms of its suitability for a ledger subject to consensus, and also will have future value once blockchain processing needs to be partitioned across multiple CPU cores or physical servers. As I argued in recent blog post, a bitcoin-style blockchain can be seen as distributed version of database multiversion concurrency control.
So I think there is sense in using the bitcoin transactional model for multi-asset transactions, to create something that looks quite a lot like colored coins, i.e. where the asset quantities sit inside the outputs of transactions. But asset quantities can and should be tracked and verified by every node. This in turns enables SPV.