For the record, cfbot only started running the recovery tests on Windows a couple of weeks ago (when the new improved .cirrus.yml landed in the tree). I don't know if it's significant that Pavel's patch is failing every time:
Previous redness was due to compiler warnings in previous patch versions.
Furthermore, I've sent an updated contrib patch with very minor improvements (completely unrelated to stream replication), and now test passes.
I suppose there's some intermittent problem with windows cfbot infrastructure which arise randomly sometimes and affects some random patches being tested. I have feeling that all patches pushed around yesterday's morning were affected and now everything works good.
It's pity I still have no idea about the source of the problems besides my look at different cfbot behavior sometimes.