How to overcome the distributed blockchain performance problem

I've found yet another naive question about why current blockchains are slow, and if any blockchain can ever be fast and decentralised. Here's my response:

Any blockchain? Yes. Will it match any of the currently hyped definitions of blockchain decentralisation, and possibly privacy? Probably not.

Here's a back-of-the-envelope idea that could work today, but is basically a clone of the existing banking infrastructure, so noone in the blockchain space will touch it:

• Introduce asymmetry between merchant accounts and regular accounts
• Introduce payment guarantors / deposit processors, linked to accounts
• Each payment guarantor gets its own side-chain (with local / basically infinite write scalability)
• Each merchant gets its own side-chain (infinite writes)
• Side-chains are PoA for the guarantors and merchants. Guarantors can pick accounts and merchants they want to do business with, based on reputation or other externalities.
• Before a payment is made, the buyer needs to top-up a pre-paid account with the guarantor, on L1/L2
• On payment, the merchant checks with the guarantor if they have enough cash off-chain, settles the payment off-chain, each side writes a signed PoA record to their local side-chain
• The side-chains are periodically settled or checkpointed with L1/L2

tl;dr: transactions actually happen off-chain, the on-chain data is sharded per-merchant (that's what side-chains are for), so can be as fast as the servers can handle them.

But good luck getting that accepted :D

(The payment guarantors in this schema basically have the same role that the credit card companies do in fiat.)