I changed nothing, but it is working today.
I imagine what is happening is that the upstream storage drops connections occasionally and cloudron simply passes that on to the user instead of having a retry loop in the background.
So, I still believe this functionality would benefit from being more robust, as before it would manage 20-50MB per go before drop, and if the thing was in a retry loop it would be able to get the archive out by just resuming downloads.
I realise it is annoying having to deal with people who insist on unreliable storage... but it is what we have.