I don’t think it’s inconsistent. My understanding is that the bad comments were in private messages and although they don’t disown then or apologise for their views, they are not making transphobic comments in threads.

In the feddit case, transphobia discourse was active and unmoderated. I have no doubt that if that was the case for .ml, then blahaj would defederate.

There is no point in federation of you defederate from every instance with bad users. Ban the user, not the instance. In the case of an instance not moderating hateful content, ban the instance.