The slight difference in the filesystem timestamp isn't surprising, because that would depend on how exactly the camera's internal operating system writes to both devices at once and the structure of the data pipelines, as well as possible differences in the card speeds. But the differing file sizes seems like a mystery.

Turns out there is a difference in the metadata. I found this forum discussion (about the D7200, but I bet the situation is the same), where people did some digging and discovered that the camera records which slot was used. Although it's just a one-bit change, with compression, this would also account for a difference in file size.

To address this, I can think of several approaches. A simple way is to rename all files using their internal timestamp (rather than the filesystem timestamp). Make sure to use the subsecond time, because otherwise if you've taken more than one frame per second they'll overlap:

exiftool -d %Y%m%d%H%M%S "-filename<${datetimeoriginal}${subsectimeoriginal}0.%e" *.NEF

Or, you could extract the preview JPEG from each image and checksum that — that should show you what's identical without regard to metadata.