The idea of the common centroid is to eliminate the effects of gradients across the two transistors. To minimize the gradient effect you want to use common centroid and have a compact & modular (square) layout. It seems that your idea can work but you might end up with a layout to long in one direction, defeating the purpose of the common centroid. You should try first by using larger w for the layout: 15u, 7.5u, 6u, and you can even combine common centorid with interdigitated fingers. Having a small unit transistors may force to have too much area overhead because of the routing.
Consider your matching requirements, if this a differential pair? current mirror? Based on that evaluate how much of matching techniques you should use.