Postgres’s PREPARE is per-connection so it’s pretty limited, and then connection poolers enter the fray and often can’t track SQL-level prepares.

And then the issue is not dissimilar to Postgres’s planner issues.