So some conclusions about Ethereum’s high level architecture.
Ethereum’s main goal is not to optimize computational efficiency, but to enable distributed and trustless computation.
Every node on the network has to run the same computations, so Ethereum is redundantly parallel.
And this is all to efficiently reach consensus on the system state without needing trusted third parties.
And because contract executions are redundantly replicated across all nodes, execution is expensive.
For example, you shouldn’t be training machine learning models or doing anything else computationally expensive directly on smart contracts.
So hopefully this creates an incentive not to use the blockchain for computation that can be done off chain.