Your colleague might have heard of the CLEF corpus of medical records, it is mentioned in a couple of papers [e.g. 1]. CLEF is a UK Medical Research Council project [2, 3]. The corpus consists of the medical records of deceased patients from a UK cancer hospital. The corpus was established under an ethics committee approval. The fact that the records are from deceased patient made it easier to get the ethics committee approval. However, the approval does not allow the corpus to be used outside of the CLEF project consortium.
There has been some discussion about how the corpus might be made available to others, in some controlled fashion, once the CLEF project ends. This would, however, require further ethics committee discussions and approval, and is likely to take some time.
If you hear of other such corpora, I would be most interested to know of them. As you say, such corpora are very hard to come by.
 The CLEF Corpus: Semantic Annotation of Clinical Text. Angus Roberts, Robert Gaizauskas, Mark Hepple et al. Proceedings of the 2007 American Medical Informatics Association Annual Symposium (AMIA 2007). To appear. (let me know if you want a copy)
John Aberdeen wrote:
> I'm seeking English-language text corpora of medical records.
> Typically medical records are unavailable to researchers due to
> privacy laws. However, medical records of deceased individuals are
> sometimes available without restriction. One of my colleagues
> believes she has heard of a corpus originating in the UK consisting
> of the medical records of deceased persons, but does not know the
> details. Does anyone know about this corpus, or similar corpora of
> medical records that are available for research purposes?
> Corpora mailing list
> Corpora at uib.no