Sharing published data?

A letter in the recent issue of Science [VOL 326 23 OCTOBER 2009] titled "The Antidote to Bias in Research" by Allison mentioned the Nature Genetics article on the issue of wide-spread non-repeatability of published microarray gene expression studies. It also touched on another recent study titled "Empirical Study of Data Sharing by Authors Publishing in PLoS Journals" by Savage and Vickers [PLoS ONE 4(9): e7078]:

In another study, refusal to share data despite policies requiring sharing was nearly ubiquitous among authors publishing in Public Library of Science journals.

I then read the Savage and Vickers article, and was surprised to find that only one author out of 10 sent them an original data set. Therefore, the authors wrote:

In conclusion, our findings suggest that explicit journal policies requiring data sharing do not lead to authors making their data sets available to independent investigators.

I can understand why the repeatability rate is so low from the Nature Genetics paper given the so many details associated with a published table or figure. However, it is a bit hard to explain why it is so difficult to share just the original dataset associated with a published work. I would imagine that the authors would be honored that others care about their publications, and contact them directly.

In my experience, I have also been simply ignored frequently for clarifications of some details in published work, or requests for some datasets. The highest successful rate is asking for PDF represents from corresponding authors.

If the basic principles of scientific publications are strictly enforced, even just by the big journals (so many, nowadays), a lot of unfounded big claims would be gone or can be easily seen through. Well, I know that's only possible in an ideal world.