In order to prime the ParserCache with Parsoid output for use by VE, we need to trigger a parse on every edit. The intended mechanism for this is by triggering a job from DerivedPageDataUpdater, see T322427.
However, we are already performing a parsoid parse on every edit: it's triggered vis RESTbase, by calling the page/html REST endpoint exposed by Parsoid (not the one in MW core). That endpoints relies on ParsoidHandler::wt2html.
If ParsoidHandler::wt2html was to write to (and read from) the ParserCache, we could avoid running the parse twice.
IDEA: Make the parser cache write probabilistic, with a configurable ratio. So we can have e.g. 5% of all edits write to the parser cache via wt2html. This way, we can sollect statistics across all wikis and page types, so we can predict how much capacity we will need to support 100%.