The backwards transformation should only help a little, only because it saves a single step that applies the inverse transfomation (the anat to epi transformation is always computed first regardless of whether you have chose epi2anat or the default anat2epi). Based on your posted command, I would suggest the following for speed improvements :

1. Use -child_epi to specify all other EPI datasets and choose a single EPI dataset as the reference EPI "parent" dataset. All EPI datasets will be registered to the selected epi_base sub-brick of the parent EPI dataset. It would certainly take much longer to align each EPI separately to an anatomical dataset than to do motion correction to a single EPI dataset sub-brick base. If your runs are within the same scanning session, then this would be preferred anyway.

2. Resample anatomical data to a lower resolution. If they have very fine resolution (0.5 mm^3), you might consider resampling to 1 mm^3. That will make the alignment and all the other steps go much faster partly because of lower disk I/O. Use the 3dresample command to do this.

3. Change the brain-stripping option for the EPI data to 3dAutomask. 3dSkullstrip is slower than 3dAutomask, but it is not as well tested for this alignment method. Use the option "-epi_strip 3dAutomask" for align_epi_anat.py.

4. Turn off unnecessary processing. Several steps may not be necessary, depending on your data. For example, if you are not interested in motion correction here, turn it off with
"-volreg off". If the data is not oblique or you want to ignore obliquity, use "-deoblique off". Slice timing correction can be turned off with "-tshift off". A resampling step can be skipped with "-resample off". Turn off all preprocessing with "-prep_off"

5. Finally, make sure the access to the data is as fast as possible. Network server access may be reduced by a variety of causes. USB drives don't deal well with large amounts of data processing either. It may be faster to copy the data locally, process the data, and then copy the data back out again.