Ahmet,
Yes, this is certainly possible: you do need to make up for it with sample size and you cannot run respondent-level models, of course, but it's a perfectly viable way of collecting data for an aggregate analysis.
I don't know that anyone's written a paper about it, because it seems non-controversial, at least in its limitations. I've heard some controversy about its advantages - I know there was one company that did most of their conjoint studies this way, thinking that only the first choice was of enough quality for building models, but many other folks think that the first couple of questions contain a lot of learning on the part of respondents and that the first question might be particularly inappropriate for building models upon.
