That's not much of an argument. I might as well say "people only think the Earth revolves around the Sun because it's the most popular opinion."
I'm not arguing with you, I'm patronizing. I'd only be arguing if I thought you would require you to be willing to substnatiate your point of view with data, or at least some form of support. However, since I do follow a tit-for-tat strategy, feel free to offer some to what follows below.
There are two questions. The first is "why those", and the second "why any". The reason why those groupings are used is because they are popular, not because they are right. Your agreement does not validate or invalidate the groupings. You are welcome to support your opinion, but your opinion on their validity is not support. You are more than welcome, as above, to provide the data on why Kiersey decided to used those traits for his groupings. I think you will find his theory published around the net.
As for "why any would be as valid"... The 4 traits in MBTI measures a different axis. Since each measures independent factors (and has been validated to do so), the "greater" differences to any combination of them is a subjective interpretation based on how you define your measurements of difference. If what you measure is included in the groupings that you proscribe to, not surprisingly, you see those differences.
Therefore groupings serve a particular purpose - they measure something specific, not "differences". It's like expressing a 3d axis in three groupings of 2 axis notation. The difference between any two of those axises is dependent on the point you wish to measure. If you select a point that flattens to extremes against two axises, it's not difficult to say that they measure "more difference". An example for the NT/NF and SP/SJ divide would be intelligence (since kiersey treed from N/S, the strong relationship to it overrides the lack of correlation of T/F and P/J). However, it'd fall apart if you were to measure "Rank in corporations and military", or "OCD", or "athletic ability", or "marriage success", or "sex".
That even ignores that expressed behaviour is roughly normally distributed. Hell,less than 50% of the people who take MBTI would select the same type when taken a year apart... So when you can't categorize the majority of your observations against any measurement type. IQ is a good example of that bias as well. Due to the distribution of S and N, Ns have a much smaller range and a large amount of Ss overlap the Ns. Yet, observationally, this is difficult to differentiate this easily (never mind have issues with rating IQ).
Alternatively, it could be argued from functions and dominance, but since the design of MBTI is to determine attitude from the P/J divide, including it only in half of them would be essentially meaningless - you couldn't relate it to functions in this design. Since Kiersey treed out the functions, starting with N/S as the primary divide, he is setting his standard of measurement against it. By choosing a different axis for the 2ndary correlation, he is simply manipulating perception - using two different measuring schemes, as in expressing magnitude in xy and xz axises. Misleading, and yes, heavily biased towards a single axis.