when your test is so poorly-designed that you can't even be certain whether you're testing a concept that exists, and even given that you are, whether your test is testing it, the test is crap even if it's not biased against the people you purport to study.

It's not my test, and I can't comment on the certainty of people who devised it / administer it today, whose opinions I suspect are more informed than mine.

Do you think that you could still be confused by "Sally thinks that Harry thinks that Sally thinks that I think that Sally thinks that whatever" or something similar?

If spoken too quickly, sure. If the test were written (or spoken slowly), I think I would give the right answer to 6th order questions at least 90% of the time.

Here's a critique of it: http://www.wrongplanet.net/postp3314609.html#3314609

The paper they link to (here) doesn't seem to be as strong as they present it in the post. I certainly agree that Baron-Cohen's claim that ToM can't be learned sounds wrong, unless he's arguing about brain structure rather than performance (that is, they can learn how to answer the questions correctly but never as easily as a neurotypical).

I also followed the citation trail to come across this paper, which included picture-based tests. An example: A green apple was placed in front of the subject and they were given a green marker (with red ink). They drew the apple someplace they couldn't see, and then the researcher put an identical red apple next to the green apple, then showed them their drawing, and asked "Which of these apples were you trying to draw?" and "When X enters the room, which apple will they think you were trying to draw?"

They tested normal 4 year-olds and deaf or autistic children (5 to 13, average age 9) on the false drawing task and a standard false belief task (what's in the box? Not what's on the label! What will X think is in the box? What did you think was in the box before I opened it?). The normals mostly passed the standard test and mostly failed the false drawing task; the deaf or autistic mostly failed the standard test and mostly passed the false drawing task. (Normal children of age average 9 were not tested; I presume they would mostly pass both tests.)

I now have a much better idea of what a non-verbal false belief test would look like, but I still think both varieties of test are useful at identifying ToM delays / deficiencies. That the normal 4 year olds do poorly on the pictorial false-belief tests suggests to me that it also is not just testing ToM, but something else as well.