I can think of multiple approaches that would work, but my inclination here would be to filter those two rows out of the dataset and overlay them in a separate table. If you are going to allow things like column resizing and reordering, then you will need correlate the column model information between the two tables to make it look seamless.