It is not a reliable resource for the evaluation of camera performance in the real world. This has been debated in detail on FM by individuals that have far more knowledge than me when it comes to the technical merit of their methods of evaluation.

For DXO Mark to conduct a significantly 'better' test than what most review sites already do would require them to do two things - spend money, and spend even more money. 1. They would have to perform statistical tests with a large number of bodies (there are 3 numbers I'd want quoted at the very least - the mean, the std dev of the mean and the std deviation). The number of bodies they'd have to use is a function of the test (pixel count would require one body while actual resolution would require multiple bodies). 2. In addition, they'd have to perform better tests. Simply taking a picture of a multiple color resolution chart is insufficient. The MTF depends on a very large number of parameters such as distance to subject, temperature, humidity (you don't think glass is a solid, do you?).
If they were doing all of this, we'd see two things - they'd show error bars and they'd claim - we get this figure by analyzing so-and-so many bodies. Given that we don't see this, the conclusion is that their 'tests' are no scientific than say DPReview's 'tests'. So take their numbers with a large grain of salt.

Plenty of errors and omissions on DxOMark - for example they suggest that the Nikon D300s will do 7 fps at 14 bit. Not a chance.

They also state that the Nikon D200 has better DR than the Canon 40D - I've owned both, and the 40D has much, much better DR.

The problem with this site is that people credit it as some sort of objective arbiter of things like image quality.

The fact that the site states otherwise is either ignored or not understood, and as a result its findings are quoted ad nauseam as "proof" that camera a is better than camera b when in fact not only do their tests fail to prove anything of the sort, but there are fundamental questions about the logic and fairness of their methodologies (there are distinct embedded biases in their testing methodologies, like the fact that they ignore the effect of "cooked" RAW files, which is why the Nikon D90 gets such a high rating for SNR) - and that's before factoring in the inaccuracies that can be found throughout the site.

If they can't get basic spec. details right, why would anyone trust the results of their "objective" tests?