Things are moving in that direction. Some of the top statistics journals encourage authors to put their papers on ArXiv, for example.[1] Creating "new elite journals" takes much more time than you'd think, though. I work in economics, and I'm aware of two new open access journals that are very good but definitely not comparable to Science or Nature (or, really, the equivalents of Science and Nature in econ). One is five years old and the other is ten, and they've probably got as high a reputation now as they're likely to ever have.

[1]: http://www.imstat.org/publications/eaccess.htm