It is actually not a bug. If you read the cohesive blocking algorithm in the Moody-White paper (
http://www.chssp.columbia.edu/events/documents/MoodyandWhite.pdf), Appendix A, then this will be apparent.
In the cohesive blocking algorithm, you find all minimal vertex separators of the graph, and the remove _all_ of them to identify the (more) cohesive subgroups. What we do is, we remove all cut vertices, then find the connected components in the remainder, and add back the cut vertices to all components we found. (Because the paper says they should belong to "both sides of the cut", it however does not specify what happens if the graph has multiple minimal separators and/or it falls apart to many components and we don't just have "both" sides, but many sides. We interpreted "both" as "all" in this case.
So, in summary, the description of the algorithm is not quite clear. So if you think our interpretation is wrong, I can be convinced. :)
Gabor