@solubrew what you describe sounds similar to Prism. It aims to i) decrease latency and ii) improve throughput. It achieves this by breaking up the single PoW problem into N problems with N subgroups mining on them and by pipelining the transaction data to consensus nodes to avoid long propagation delays upon solving the PoW puzzle. The nice thing is that Prism uses a sortition mechanism to determine which sub-problem the miner has solved, so one miner can’t easily choose which sub-chain to mine.
I don’t follow why you’d want to only allow one tx per address, but this is something that has been suggested as a DoS protection for Account Abstraction. Might be worth checking out for inspiration.