The heatmap scale argument says
"character indicating if the values should be centered and scaled in either the row direction or the column direction, or none. The default is "row" if symm false, and "none" otherwise."
Looking at the function, it appears that if scale is chosen (either row or column) the scaling is only done *after* the hierarchical clustering is performed. So the scaling from the 'scale' argument is only for choosing the color scale, not for clustering the data.
While the documentation is not technically incorrect, since the documentation doesn't say specifically either way, it seems to at least be an important aspect to document and I don't see it anywhere.
Personally, I think this is an odd choice, since I would want the color image to visualize the same thing that the clustering algorithm used in making the dendrogram. But I'm sure others would disagree.