We do NOT want to embark on a third (or fourth or fifth depending on how you count them) portal implementation!
The current implementation isn't bad, the rewrite's a decent reorganization of the Tcl API (logical use of namespaces) and simplification of the datamodel (making things not objects that don't need to be objects)
Other than that it's very much the .LRN new-portals system.
The current "portal" source in HEAD installs (or did when I last had time to work on it) for both PG and Oracle.
I've very recently added some Tcl API in .LRN new-portal that lets you define your datasources in your Tcl APM callback rather than in SQL, meaning you don't need to write both Oracle and PG definitions. Simplifies things somewhat. I'll move that to "portal" soon.
The only thing that really needs to be done to boost performance significantly is to add caching ... and I don't see any indication that the portal system performs poorly.
Certain *portals* in .LRN are expensive but that's not the fault of the portal system itself.