Thomas dcm2niix assumes that the DICOM images are truthful. Note that for the Siemens image you refer to, dcm2niix is faithfully converting the MosaicRefAcqTimes tag from the Siemens CSA header to seconds. You can see this by viewing the DICOM with a hex editor:
DICOM files often exhibit small errors due to rounding errors, particularly for variables stored as text. However, sequences also have small variations added in to pad adjustments. A nice example of this is the Philips slice timing gap due to Prospective Motion Correction. Multi-band sequences add an extra layer of complication. The variability you have spotted is negligible with regards to the sluggish hemodynamic response we are trying to compensate for with slice timing correction, so I think you have permission to ignore. However, feel free to follow this up with the Siemens Research Collaboration Manager associated with your center (though they may note that the example you provide was an early CMRR research sequence, so you should look to see if you see similar variability in Siemens product sequences).