Addressing the idea of validating computations with experiments, this is a practice that is done constantly in the scientific community. In fact, in general a CFD solution can never be truly validated without either a direct comparison with either an analytical solution (very, very rare) or experiments. Naturally, experiments are the typical route here when possible.
Lately, Physics of Fluids has had a lot of work on Rayleigh-Bénard convection. Just browse through recent issues of that and you ought to find a pretty nice chunk of work which will, of course, reference other, more canonical works in the field.
http://pof.aip.org/search?key=PHFLE6&societykey=AIP&coden=PHFLE6&q=Rayleigh-B%C3%A9nard&displayid=AIP&sortby=newestdate&faceted=faceted&sortby=newestdate&CP_Style=false&alias=&searchzone=2
The thing about that particular phenomenon is that, at least today, there is not going to be a lot of experimental data on simple Rayleigh-Bénard convection because it is already a well-studied problem and in many cases it has an analytical solution for the interesting bits of the phenomenon such as the hydrodynamic stability. You may have better luck finding data for variations of the traditional Rayleigh-Bénard problem.
The other thing to keep in mind is that quite often, you will see a numerical work in a separate paper from its experimental counterpart, especially because oftentimes one inspires the other.
I will also add that during my undergrad, FEA and CFD courses were offered by my university as electives that attempted to do a mix of the background theory (e.g. Galerkin methods) while also providing familiarity with various software (generally ANSYS and Fluent). Of course it is a class so, as Vadar2012 pointed out, it will still never quite be the same as what you learn from reading a couple papers and then doing it. Then again, most undergraduates don't have that opportunity so the class is at least a good start.