Yes, it's very common, but that doesn't make it better ;-)
1) there are always some differences between templates, e.g. if you preprocess your data based on a non-linear MNI template version vs. a linear version (with less spatial accuracy) then you will receive somewhat different results during normalisation.
2) often figures do not just serve for display purpose, the authors might infer something, e.g. that a cluster is located in a particular branch of a sulcus or some subregion within the laterobasal group of the amygdala. If you look at the normalised data from the study one can expect much more blurring though, so while the coordinates might agree well with a particular sulcus in a particular brain region your data might be much less precise. Thus it would be helpful to display the data on a volume reflecting your subjects (e.g. some normalized average volume or probably better, a smoothed, normalized version).