Since none of the files that are passed to mkacisrmf are aware of the date of my observation, how is the time-dependent gain included in the RMF?
The correction for the time-dependent gain was taken care of when you reprocessed the data with acis_process_events and set "apply_tgain = yes". Since the events have been corrected for the change in gain, it does not need to be accounted for a second time in the RMF. The RMFs are calibrated with respect to some fiducial time and the TGAIN that was applied to the events corrected them to that fiducial time.
So while the RMF isn't aware of the observation date, it does not pose a problem.
For more information on mkacisrmf and its calibration, read the Creating ACIS RMFs with mkacisrmf why topic.