A cursory look at the paper reveals that the author is not a professional mathematician. The algorithm looks very simple, and the bibliography is extremely short. So according to Scott Aaronson's criteria (mentioned by svenkatr), the probability that this particular approach works is small.
This is aggravated by the statement in the website that the algorithm was already obtained in 2002, but the author waited for computer simulations - this makes no sense (either the proof works or it doesn't). Now it's true that in practice people can solve most SAT problems that they want to, probably much larger than the ones mentioned in the paper. But NP-hardness is about worst-case analysis.
The real question is whether this new proposed algorithm is faster in practice on real-world instances. The author has done no comparison, and so (since people have worked on optimizing the existing approaches), his method is probably worse than the competition.