Before going on, still, we should have a look on the general structure of a phase-matching AF detection system and what key components are contained in such a system:-
Credit: http://kmp.bdimitrov.de/technology/AF/index.html
In general, as per each AF "point" that is consisted of two crossed linear sensors (which is generally known as a cross sensor by people), if a f/2.8 sensor is included, there are several possibilities as follows:-
1. Modifiying either the horizontal or the vertical sensor to a f/2.8 one. Indeed, this is the easiest and cheapest way to go. What it needs to do is to replace the pair of image projection lenses as seen in the above diagram. The purpose of this is to measure light from the outer part of the glass, instead of measuring the AF closer to the centre. There is simply no need to change anything of the AF sensor array as it is only about the AF projection lens pair that all matters!
Now with the f/2.8 sensor, if with proper software logic, higher AF accuracy can be achieved for fast lenses. What I mean for the "proper" software logic refers to the camera should choose the better measurement value that is closer matched to the actual f-number that is used. But this implementation has one major drawback, that is, with slower lenses than f/2.8, one of the crossed sensors will become blind and virtually the cross sensor will be reverted back to linear!
2. Both of the sensors are now f/2.8 - It is almost impossible as once a lens slower than f/2.8 is mounted, that AF point will be completely blind!
3. Additional AF sensor(s) *on top of* the old f/5.6 sensors - this is actually the prefered approach technically as no cross sensitivity is lost even if a slower lens is used. That will involve a new layout design of the AF sensing unit and array, though. Yet once again owing to the "stop gap" nature of the K-5II/s, this is quite unlikely.