Sounds pretty solid if it works as you've described - excited to see the finished product. It sounds like other blockchains could potentially adopt it as well. It's probably worth noting that a good custom caching system for JSON-RPC requests from nodes is often highly valuable. For Steem we wrote a redis backed custom caching layer called Jussi, it probably is still part of the Hive stack today. The idea is that the vast majority of requests are for read only data involving irreversible blocks and therefor there isn't any need to ask the node for that data - if the custom caching layer is aware of what is irreversible and what is not then it can avoid hitting the nodes with unnecessary requests. For something like a social media site this is perfect and allows you to handle millions of requests per second with only a few nodes. You make a really good point about smart contracts, however, I imagine that a lot of requests from smart contracts are also read only and only executed by the node serving the request rather than needing to be executed by every node on the network - some of these types of requests could probably be served from a cache as well.
RE: Fractally's Groundbreaking New Database Outperforms LMBDX / EOS / Hive / Chainbase / RocksDB