Do you think that some form of dynamic time warping algorithm, performed on the data of each distinct group, could help give a better overall view of that group's behavior in a static heatmap image? Presuming participants share similar scanpaths within a group.

Just a thought, I understand the data may be too noisy / have too much variance for this to work in practice.

The problem with applying complicated algorithms to an infographic is that it makes it more difficult for the reader to know what exactly they are looking at. I think infographics should always have a "transparent" quality where a reasonably seasoned person could explain exactly what each individual point of data means. When you start applying sophisticated kernels to the data, this becomes hard to do.