I am at the moment trying to understand what kind of benefit I would have by using three cameras for stereo vision rather than two cameras?
I mean, i would only have more constraints related to the fundamental matrices, but at the same time I will have multiple fundamental matrices, so it seems to outweigh it self?.. so what is the benefits?
You get redundancy. With two cameras, each depth estimate depends on the intersection of a pair of lines. When those are near parallel, this estimate can be very bad -- a small error in the data can introduce a large error in the solution. Since your data are typically discretized (at the pixel level), such small errors can be quite common.
With three cameras, you have three pairs of lines. Those give you three different position estimates, and you can average them (or even do more clever things, likek taking into account the stability of each estimate). On the downside, you have three estimates, and you have to do SOMEthing with them, and that requires thinking about stuff. With just one estimate, you have no choice. :)