I have done a few simple experiments in the past comparing CTEs like this to JOINS, but the resultant query plans were the same. CTEs seemed easier to follow when troubleshooting issues, so I left them as such. Do JOINs become better than CTEs at a certain point?
Read up on from_collapse_limit. If the query can re-write subqueries to collapse the join problem, then it will at first but then once it reaches that threshold, then it won't try anymore to avoid excessive planning time. That's when things can go awry.
