1. Negative means data that contradicts the popular hypothesis. This is frequently reported very prestigiously, though sometimes care has to be taken depending on just how deeply held the refuted belief is.
2. Negative means data that is highly inconclusive. This data is often non-reported since people continue experiments until they have a story or just drop inquiry. It should still (in some cases) be reported along with the final, telling data but probably isn't.
3. Negative means data that supports already "known" beliefs. Totally "uninteresting" results which just serve to give us the tiniest greater justification to believe what we do. Rarely reported.
I can see (2) but mostly (3) being suitable targets, but you might have a different idea in mind for what negative data is. In the end it sounds like a need to make big advances in meta-analysis in order to have data reported against a more quantified context.
Global, public, and automatic meta-analysis of the state of scientific belief is a hard problem I would love to see solved.
and to echo the above comments - arxiv is not prestigious. you wouldn't write a job market paper or pride yourself in terms of 10 publications in arxiv rather than 2 publications in Nature. i havent thought about the 'how', but i am convinced that the prejudice against negative data has to disappear for this whole thing to work.
Now the idea of being able to search a scientific question and somehow find a meta-analytical summary of all the inconclusive results ever reported would be flat magical. If the summaries were really useful it'd likely generate a large number of telling results almost automatically.
Of course, a resource like that is almost mythical sounding for today.