One of the key differences from what I argued upthread to monsterer, is that "providers" aggregate transactions and thus users can reason about malfeasance on less granular basis than every transactions for itself. This enables users to organize around "providers" which are behaving correctly.
I'm building a clearer picture of your design now. Providers aggregate transactions because users are not suited to the task of building blocks due to their ephemeral online presence. All PoW comes from users in this design, not providers. Users pick transactions from providers to place in their block when they send a transaction. In order to try and prevent a majority PoW from censoring transactions, users can look to these providers in order to tell if a particular fork is censoring transactions.
However, I see a problem: why can't the majority PoW simply create a majority of providers as well, and therefore gain a greater than 50% chance that users will miss information about censorship?