Whatever, they're fooling themselves. Even if people did test 3+ categories the same 75-90% of the time, which they probably don't, the only scientific data that gives you is their preferences on the particular questions given. You can't really extrapolate that into a conceptual model that summarizes a type's preferences without using induction, hence the inherent problem of subjective bias.
Typology is unscientific by nature and acknowledging that from the beginning is paramount.
I vote we start using NSTT, New Subjective Typology Theory. I just made that up, and I'm going to call our version of typology that from now on.