I believe the idea is to make it so this technique isn't useful for analytics anymore (edit: because your MAC mining data would be telling the user is in 100+ places at once). If it would work is an entirely different story because since the devices are stationary and not moving (like a device in your hand or pocket would) would it be easily detectable that it was a "faking" device.
It is not so much about your device being at many places at the same time (though it is a nice side-effect) as it is about polluting tracking databases. If you normally have 10 people in an area, and all of a sudden there are 2000, you know something is wrong. But there is no telling which of these 2000 devices is the real one, meaning you have to accept false metrics, or remove all of them from your system (including the real device).
Pollution is essentially what I meant apologies for my poor phrasing.
However detection of these devices could become easy considering they probably wouldn't move a lot, meaning you'd have to either (a) move the device frequently and randomly or (b) the antenna that reproduces the signal would need to add a bit of weakness to the signal to adjust and make the user look like he is moving.
(e.g. If the signal strength stays the same constantly then it might be easy to tell the user isn't moving ever and therefore likely not a spot they are at currently, or even if everyone shares a very similar signal strength).
According to the description on GitHub the system only adjusts the sequence numbers.
