So the analyze(and freeze) are done before the new cluster are fully populated. Is the time being taken maybe for the loading schema/data portion?
No, the upgrade is stuck on these 2 stages indeed, maybe 50% on the first and 50% (or a little more) on the second. When a run them outside of the pg_upgrade procedure it is still slow, so pg_upgrade is not to blame. I guess I will need to repack the biggest tables first.