The division would be acceptable but only if the insights provided by the other were included. For example, if research identifies that a specific functional form of a relationship is an "s" shape, then unless someone comes along to show that the relationships is actually a z shape, then all models should use the s. But people keep using a v shape! Poor specification does have implications, no?
Of course, if scholars interested in only one type of explanation (i.e., IR or comparative) do not use the best models, then 1) their work is wrong and/or 2) the peer review system is dysfunctional for allowing such work to be published. The problem seems to be that the two different areas largely publish in different journals and when they do publish in the same journals editors tend to select reviewers that are familiar with one strand or the other. To each their own.
The end result: we end up with a poor understanding of the topic.
Solution: ask what the best model is, what is the state of the art operationalization for a specific variable and then use it.
Can't we all just google it? Googletruth - coming soon...