yes, this is because SGP is scaling the 14-bit DSLR data to 16 bits when saving the fits files.
i asked as a feature request that SGP not do this but it was rejected. from this thread it looks like PI calls DCRAW on NEF/CR2 with different overscan handling parameters than SGP does, so it may be a moot point - even if SGP did not scale, the geometry would be incompatible with PI.
still, on first principles i think an image acquisition program should not scale the data coming off the camera.
i also think there might be some funny stuff when the data becomes saturated (meaning saturated values don’t show up as 0xFFFF in the scaled values) but i have not yet hooked up my 50D to SGP to test that out.