This isn't really a poll, though. You aren't presenting a question, or asking for a respondent to select between four answers specifically related to the question. It might show that, given a choice between four totally random positions, with no other qualifying information, a majority of the respondents would randomly choose position three. If the poll asked "What is your favorite motorcycle?", and the enumerated responses were A) Suzuki, B) Yamaha, C) Royal Enfield and D) Harley, I don't think that the truism you mentioned would hold true, even if the majority of the respondents didn't know much about motorcycles. Just based on name recognition alone, a non-rider would be more likely to choose a brand he'd actually heard of at some point in his life, which would probably negate choice "C".
Is there some study that shows that in a poll with a specific question with a choice of four responses related to the question, the majority of respondents will randomly choose "C"?
"No matter where you go, there you are."