[kijinbear]

These guys did exactly what's being proposed in the article.

http://www.win.tue.nl/hashclash/Nostradamus/

In November 2007, they posted the MD5 hash of a PDF file containing the name of the winner of the 2008 election. That of course doesn't prove that they knew who would win the election, because they prepared 12 different PDFs with 12 different names in it, all of which were crafted to generate the same MD5 hash. I don't think the hypothetical jury is going to like it.

Okay, let's try again with SHA256...

[drv]

The article mentions hashing the same data with multiple hashing algorithms; I would think this would be effective at preventing collisions (presumably different hashing algorithms are not vulnerable to the same collision attacks).