interfacing or a super-model ?

Yesterday, I forgot to mention another blog, on which prof. Smith's blog was based.
It can be found at: http://wolandscat.net/2009/10/01/the-crisis-in-e-health-standards-ii/

The latter brings me to an interesting question I am asking my self often: should we (in healthcare) try to build "supermodels" that try to encompass almost everything (as HL7 does with the RIM), or should we allow for more "isolated" models (in CDISC: SDTM, CDASH, ODM, ...) and have them use elements of each other, and build interfaces between them, as IHE does with its profiles?
I haven't got the full answer yet, but observing the immense problems e.g. HL7 encounters with the RIM (some ontologists say the RIM is basically incorrect, and there are currently already over 80 versions of it), and the monsters it produces when it comes to implementation, I am more and more thinking that the IHE approach is currently the better one. Of course it is more work (the n-to-n problem), but at least it leads to implementable systems.