An update on what I've done so far:
I factory reset the pager and cleared all user data. Then I flashed the firmware again. Then I factory reset and cleared user data again. Then I programmed it with a very basic test profile that only had 1 zone and 1 knob position scanning 1 talkgroup with no subgroups, and it was still tracking very slowly. I noticed its even slow to change between the knob positions on the factory default programming, so I think that's just how this pager is.
I turned on voice buffering, and I can't tell if it's helping recover those missing words from the beginning of a transmission, but I noticed it has another effect that I haven't seen mentioned anywhere
except here kind of. I thought voice buffering just simply delayed the voice for a second, but with voice buffering enabled, if one talkgroup is being played live and another is talking at the same time and being missed, it will sometimes store that entire missed transmission and play it once the first transmission is over. It's very hit or miss. I honestly don't know how this is possible because TDMA isn't a thing on my network and I don't think the G series pagers can tune to multiple frequencies at the same time. Hopefully I'll get it on camera, but I've seen it twice now start playing audio from a transmission that was entirely missed and already finished.