Matthew wanted SAT in Farseer Physics Engine, so he worked out a solution that would work together with the distance grid. Problem was that Farseer did not support specific shapes and thus he only implemented polygon vs. polygon collision detection. He wanted
circle vs. circle, circle vs. polygon and polygon vs. polygon implemented so he also started on getting real circles into the engine too.
It was too much work tho, you have to have 100 things in your mind at the same time, so he left the real circle implementation and started working the polygon SAT code instead. He got a solution working but it had some bugs, unfortunately he does not have
a lot of time (personal life, moving house and so on) so he has not been active for quite some time. He has worked up a new SAT implementation tho (the one you mention), but it requires more testing and still has some bugs.
Depending on the algorithm, real circles will be faster than polygon approximations and it will not have the jumping problem (hitting the edges of the circle). It calculates the contacts on the fly and thus it will also bounce correctly.
However, I just got word from Erin that Box2D is very close to be feature complete - We have a working C# port that is only 3 revisions behind the main trunk (that will change later today as I will be porting over the latest changes) and thus we will have
real circles and a brand new SAT implementation in C# very soon. I'll make sure to announce it once I think it is ready to be tested in the wild.