This is a theoretical question. I am currently studying the rendering equation and I don't understand in which case area sampling or hemisphere sampling is better and why.
Another thing I would like to know is that if the result would be better if we took an average of both methods ?
Veach & Guibas, "Optimally combining sampling techniques for Monte Carlo rendering", Proceedings of SIGGRAPH 1995.
This classic paper addresses exactly the problem you are concerned with.