A small comment on L1 scalability: by itself, moving from POW to POS doesn’t bring more scalability. You need to play on many part of the blockchain and consensus to push better performance. You can skim some of the different approaches in this list of Notable works in scaling.
Now onto your questions. You are right, when we compare L2, we have to look at many more parameters than what you have with L1 solutions. Personally, I would have a hard time comparing those solutions in depth because on one hand I haven’t read that much about the other protocols, beyond some podcasts debate, and on the other I focus on the papers from Arbitrum, which lack the implementation details. So far the best comparison resources I found are:
Broadly speaking, I feel like we are trying to escape L1 consensus, yet we are realizing that we need some consensus on the data (Note: I am still wrapping my head around the Data Availability problem). For instance, Arbitrum requires all the validators to agree while on DYDX (that use Starkware ZK Rollup), there is one entity generating the proofs. OR have the advantage that they can onboard dev and by extension much more easily: they keep L1 Composability, despite their optimistic nature. ZKR seem much powerful in the long term but have issue onboarding projects.
What does this mean for the end user then? As you pointed out, there will be many trade-offs. I am not sure how users will understand the different guarantees they have and all the means, if any, they can use to detect/stop a fraud. This brings back the question of decentralization and the trustless nature of those protocols, as also mention in @Ajibike_Jimoh’s and @Astrid_CH s comments.