If there are multiple jobs building the same maven module, currently it's more or less random which of these jobs is included in Jenkins' dependency graph.
The mechanism should be improved to choose the more relevant jobs - if possible - and the deliver predictable results.

I've implemented a prototype here:
https://github.com/kutzi/jenkins/commit/0e64a6ff3f9cb424b3f6b3adfa06ec1eb134e72c