Rhapsody does support XMI, but in my experience, XMI interchange between two vendor implementations has not proven that successful.
The real short case analysis is that we have a very cross-functional domain: business level analysis modeling, IT, and embedded -- including requirements modeling, software architecture, detailed software design, and generating specifications and documentation from the model for each of these different areas. EA succeeds in these areas and handles the different domains pretty well. Rhapsody is very centric on embedded only and promotes the code generation runtime framework which I am not interested in. I have also found that Rhapsody 5.x did not support all of UML 2.0 that I was interested in. I am unsure how 6.0 fares.
I find usability in EA to be better.
Forward/Reverse engineering supports more languages whereas Rhapsody requires you to launch separate versions of the tool for each language and modeling information can't be shared between them -- maybe this is fixed in 6.0.
Documentation in EA is far superior. The MS Word Master Doc / SubDoc is really, really useful.
EA Plugin architecture for UML Profiles, etc. is very useful.
There are other reasons, but I am out of time.
I do think EA should put executable UML on its roadmap. Rhapsody is very much executable UML driven, but this requires some kind of underlying framework, and becomes very platform-centric. Perhaps MDA can help, but I think there is quite a journey ahead.