There's not much of a theoretical difficulty, is there? You just need to scrape titles and abstracts from whatever journals' websites that you choose to include -- this seems like something a reasonably net-savvy programmer could put together rather quickly (heck, you could probably write a quick GreaseMonkey script to do most of this). Then you just need a user-base of "open-source reviewers," a catchy domain name and some web-hosting, and you're good to go.

If you're talking about the added access to articles through MathSciNet, then I think you're out of luck for the immediate future -- support the Arxiv and free online journals and hope for the best.