Hi Steven,
Yes I agree your images are not undersampled. The width of the features we are looking at (eg the Balmer lines) are much wider than the pixel size. In this case, as the normal image was also with the sub exposures aligned, the algorithm is enhancing the image contrast in a way that I would not have expected it to do and whatever the algorithm is doing must be suspect so probably best avoided for scientific imaging.
One test would be to measure the Equivalent Width of the lines in the two spectra. (Perhaps better done on a reference star spectrum with relatively sharp lines and good SNR) This should be invariant under any procedure (Indeed it should be independent of the instrument) ie as the maximum intensity in the line increases, the width of the line should decrease to maintain the area under the line.
(Note that variations in resolution along the spectrum are to be expected with the simple grating in the converging beam setup. The focus varies due to field curvature so only one wavelength can be in focus and chromatic coma increases with wavelength so that for example if the short wavelength end is in best focus, features towards the red end will be increasingly blurred.)
There is a case for using drizzle in undersampled slitless spectroscopy however eg as used in the WFC grism spectrographs on the WFC on the HST
http://adsabs.harvard.edu/abs/2006hstc.conf...85K
and planned for the JWST
https://ntrs.nasa.gov/archive/nasa/c...0150023515.pdf
Cheers
Robin