I needed a way to load up the children of a very large table with decent referential integrity a lot faster than joins and sub-selects. What I came up with was a stored procedure that loops over the parent ids, and loads up the results into a temp table.

fast_child_loader.sql

create or replace function locations_for_demand_set(ds_id uuid)
returns table(id                 uuid,
              demand_id          uuid,
              is_anchor_zip      boolean,
              latitude           double precision,
              longitude          double precision,
              name               varchar,
              subdivision_rank   varchar,
              type               varchar,
              division_permalink varchar,
              zip                varchar) as $body$
declare
  rec  demands%rowtype;
begin
  create temp table answer on commit drop as
    select * from locations
    with no data;

  for rec in select * from demands where demand_set_id = ds_id
  loop
    insert into answer
      select *
        from locations l
       where l.demand_id = rec.id;
  end loop;

  return query select * from answer;
end
$body$ language plpgsql