Hi Jacob,

QuickNII allows the user practically to find a suitable rectangle in the atlas which covers an entire image, with all of its squeezing and other possible distortions. Because of this, it's very unlikely that the identified rectangular atlas portions for a series of images are going to have the same dimensions. If the image series comes from block-face imaging, the chances are higher, but still not 100% sure. I didn't want to mislead users that there are tons of pixels/voxels in the atlas with infinite resolution, so the exported cuts have the actual dimensions in pixels what the identified 3D rectangles have in voxels. A practical detail is that this way the exported images are small. 
There is some information about how image pixels and atlas voxels relate to each other here: https://www.nitrc.org/plugins/mwiki/index.php?title=quicknii:Image_coordinates
And while they haven't made yet into to Wiki, there are some practical ideas in neighboring topics, like how one can manually or programmatically overlay the resulting atlas cut on the actual image: https://www.nitrc.org/forum/forum.php?thread_id=12559&forum_id=9082 (manual) and https://www.nitrc.org/forum/forum.php?th... (Python)
For building volumes another direction is available though: creating a sparse NIfTI file from an aligned series, which can then be viewed together with the standard segmentation files in ITK Snap or similar tools. The example code for that is Java and it's available here: https://github.com/Tevemadar/QuickNII-extras/tree/master/Java

Best regards,
Gergely