The hydrogen bonding in meso-hydrobenzoin requires a disfavorable steric interaction between the two phenyl groups. The hydrogen bonding in racemic hydrobenzoin (which is the picture posted above) does not have these interactions when the hydrogen bond is invoked. Those steric interactions have to be worth something, right? The hydrogen bond itself is essentially a wash, but the steric interaction makes the meso form higher in energy. I think that is the best explanation for why the racemic product is lower in energy than the meso product.
The product distribution, however, is not determined by the most stable product in this case, but by the lowest energy transition state. You would get the thermodynamically more stable (racemic) product if the reaction were reversible. However, hydride delivery from a borohydride reducing agent is an irreversible process so the lower energy transition state is what dominates.
You might look at the following articals for examples of chelation controlled additions:
Cram, J. Am. Chem. Soc. 1959, 81, 2748.
Nakata, Tetrahedron Lett. 1983, 24, 2653 and 2661.