It seems you're saying that a categorization system that doesn't describe its subjects in complete, full, everlasting detail is worthless. I'd have to disagree.
Look at the poll right now, or its predecessor. INTP has a higher tally than all the other types, which is particularly interesting considering its relative rarity in the general population. You could therefore conclude that an INTP is more likely to vote in a poll on HN than, say, an ISFJ. That's information. Useless information perhaps, but information nonetheless. If MBTI didn't actually indicate anything, then the tallies would more closely fit the type distribution in the general population.
Your concerns about the test's reliability are valid, but any attempt to categorize something that we don't completely understand is going to have its inconsistencies. It's a far leap from there to useless.