Good questions!
It's tricky to validate the accuracy of the simulations considering that there is no experimental technique that can do the exact same thing MD (molecular dynamics) does, in the sense of time and space resolution.
However we can and have validated our results on various experimental data such as protein crystal structures, NMR data (see Nate's publication) and various other experiments.

It has been shown in various papers that it is possible to fold proteins with MD simulations to their native structure seen by Xray crystallography. However as MD simulations are still quite slow, we can only validate this on relatively small and fast-folding proteins. So for all we know, the current MD force-fields might for example not be able to fold huge proteins.

Noelia is also about to publish a paper where she does protein-ligand simulations and then validates the results of the simulations to experimental results with quite good success.

So in general I would say that over the last years the accuracy of the simulations has increased a lot and can be used to partially replace some experiments, or at least be used as an extra tool in drug design, which is also what we are trying to sell to the public :)

What Chilean said. I meant "selling to" as "convincing someone". But not just crunchers for our project. We also want to convince the industry that it can be useful for real drug design applications. Most pharma companies are still a bit behind computationally, or at least are being a bit conservative about entering this field.