My current compromise approach is pretty crude: for everyday small and frequent interactions, just use L2; only do two things on the mainnet—deposit and withdraw large amounts, and modify key permissions. Basically, the mainnet is like a safe, while L2 is like a change purse. The experience is smooth, but you have to accept the underlying dependencies of bridges, ordering, and upgrade keys.

Recently, after some mainstream public chain upgrades/maintenance, the group started guessing whether the project would migrate. I instead drew a diagram of "who depends on whom": if a dApp's core liquidity, oracle, and cross-chain bridges are all tied to the same chain, it claims to be able to migrate at any time, but the actual cost is ridiculously high... Conversely, contracts designed for multi-chain, with front-ends capable of switching routes, are truly "migratable."

Don't get too hung up on gas fees either; batch when possible, sign fewer transactions, and avoid giving three or four permissions with a single transfer. The experience savings may not necessarily be money, but less hassle and fewer pitfalls. Anyway, I’ll stick with this for now.