I wouldn't be surprised if COI encounters issues with automated abuse detection and spam filters in those servers. When you're in an active conversation, the client is rapidly sending 1-line emails with strange headers.