Pew! Pew! Pew! And another straw man crumples to the ground...
The notion that the MBTI claims to assign people to "pure" all-or-nothing categories is probably the silliest of the memes that regularly recur in MBTI debunkings, and it has the dual charm of being bothan inaccurate characterization of the MBTI and — in its misplaced emphasis on the shape of the distribution curve — a red herring.
Nobody knows for sure at this point but, as I understand it, the existing studies suggest that it's likely that most or all of the MBTI dimensions — like the four Big Five dimensions they basically correspond with — exhibit something like a normal distribution, with substantially more people near (or in) the middle than near the extremes. For what it's worth, Jung thought more people were essentially in the middle on E/I than were significantly extraverted or introverted, and Myers allowed for the possibility of middleness on all four dimensions — so the in-the-middle possibility really goes all the way back to the MBTI's roots.
Myers believed that it might turn out that one or more of the dichotomies was truly bimodal to one degree or another — with, in effect, a more or less empty (if narrow) zone in the exact middle of the spectrum. But she never asserted that that theoretical possibility had been factually established by any respectable body of evidence, and the 1985 MBTI Manual (which she co-authored) stressed that the evidence for bimodality was sketchy at best. And since then, as I've said, quite a lot of evidence has accumulated that seems to suggest that most or all of the MBTI dimensions exhibit something more like a normal distribution.
In at least one of the early versions of the MBTI, it was possible to get an "x" on any dimension. The current version assigns people a (tentative) type on each dimension, but that's a very different thing from saying that it isn't possible for someone not to have a preference — and the MBTI Manual specifically notes that someone with a score near the middle is someone who has essentially "split the vote" rather than offered much evidence of a preference.
The "Step II" version of the MBTI includes five "facets" for each dimension — just as the NEO-PI-R has six facets for each Big Five dimension — and allows for the possibility of being, for example, on the T side of three of the facets and the F side of the other two.
More importantly, I'd say, there was really no doubt in either Jung's or Myers' minds that people on either side of the dimensions fell along a notably wide spectrum from mild to strong preferences. So, regardless of where anybody wants to come down on the "exact middle" possibility, if they take the position that, e.g., all introverts are equally introverted, their perspective is way out of line with Jung, Myers and every respectable MBTI source I've ever encountered.
As a final note: At this point nobody really knows how close to the middle how many people are on the MBTI (and Big Five) dimensions, because the current state of both the MBTI and Big Five is such that it really isn't possible to determine exactly where anybody falls along whatever the real, underlying (and substantially genetic) spectrums may be. So it seems to me that anybody who thinks that the existing data on either the Big Five or MBTI has clearly established the shape of the distribution curves is very much overestimating the ability of the existing tests to accurately quantifystrengths of preferences.
But the main point to keep in mind is that, at the end of the day, the worth of the MBTI and Big Five is mostly going to hinge on how good a job those typologies do in nailing down what personality-related characteristics tend to be associated with the corresponding preferences, and not on how many people turn out to be at any particular point on any of the relevant spectrums. And in any case, the MBTI certainly doesn't stand or fall depending on whether any of its dimensions exhibit a "bimodal" distribution.