Large/complex type diagrams do not usually carry their weight, IMO. There are many good uses of a multitable such as [anagrams](https://forum.nim-lang.org/t/5917). As long as the history of an instance does not use `add`, `allItems`, or `dups=true` then it's pretty clear to client code they have no dups. They are free to use `seq[T]` values when they need/want a more full-featured API.
"State of the art" gets over-used, IMO. I don't do anything in `lptabz` that could not have been done in the 1980s. Fast, flexible engineering is sadly too rare. While I try my best to keep `std/tables` from straying into cache/branch predictor/etc.-hostile territory, your question/proposal is really for Araq. It's not up to me. I have not yet added the abilities of `sharedtables` to `lptabz` which would be one blocker. Plus many tests. I did [open an RFC about this](https://github.com/nim-lang/RFCs/issues/201), though we closed it after adding a better integer hash.
