I'm not an expert in neuroimaging, so I had to search a little bit to learn about how carry-over designs apply to fMRI (I found Aguirre, 2007). Thus, feel free to correct me if there is something specific about this problem domain that influences the correct answer to this question.
However, based on general principles of experimental design of repeated measures experiments, you always have to choose
- how many factors you want,
- how many levels you want for each factor,
- what levels you have for each factor, and
- what factors you hold constant.
The particular levels you choose are directly tied to your research question. There are often trade-offs between the cost of resources required to have more experimentally manipulated levels and the benefit of additional questions that can be answered. Thus, I don't think a black and white answer can be given. You need to weigh up the pros and cons.
And presumably you have already made many experimental design decisions to get to this point. There's nothing magical about 3 levels to a factor. If anything, in experimental psychology two levels seems more common. And you could have four or five or more.
- Aguirre, G.K. (2007). Continuous carry-over designs for fMRI. Neuroimage, 35, 1480-1494. HTML