Yes, you are correct that the computational of running the smart contracts is very small in relation to the proof of work.

But actually, the main cost is not computation, but storage. Storage (any variable that is holding data inside the contracts) is extremely expensive because there is very little of it, and has to be persisted on the blockchain forever.

In fact, view functions that don't change the state of the blockchain can be executed for free at any time.

Related to that, people have started using old procedural generation tricks to store "large" amounts of data on-chain. For example, as far as I understand it, Aavegotchi [1] stores base layer SVGs, metadata and functions to use these and render the actual end-product at view-time.

Blockchain constraints are eerily reminiscent of the demoscene [2][3][4] (just using the opportunity to share stuff I like, don't mind it).

[1]: https://github.com/aavegotchi/aavegotchi-contracts

[2]: https://www.youtube.com/watch?v=bD1wWY1YD-M

[3]: https://www.youtube.com/watch?v=qJegM2UgEZU

[4]: https://www.youtube.com/watch?v=0w_xEUoK79o

This: If you don't have some noteworthy transaction cost, the network gets flooded with spam past its other scalability limits.

You can shrink it somewhat in absolute terms, but you need to throttle them somehow.

Yes, but you also need to store and compute over a massive global state, and any execution is replicated across every full node on the network. Too cheap = DoS.

The proof of work is unrelated to smart contracts or transactions in general. Even if no one were using Ethereum at all today, the proof of work done would be the same as if it were used to capacity. The proof of work merely protects the consensus: it is not a cost of use.

consensus on the distant past isn't very interesting. The proof of work is like a data center, or by sharing a whole cow's meat -- it costs almost all or nothing, not linear in usage, but the cost is borne by some method of division across the contents.