I understand why that could be true, but I also wonder if Google did that on purpose a little bit, for the same reason they make you configure their Wi-Fi routers and other devices over the internet.
There are a couple of legitimate reasons that the on-device model is lower accuracy: compute and size. Models are often reduced in complexity when running on edge devices, but cloud-based GPU (TPU in Google's case) models can require far more calculations in order to squeeze that bit of accuracy. As for model size, a 100MB model in the cloud is not a big deal, but having to download a 100MB model on every edge device is expensive in both time and bandwidth.
Outside of these reasons, Google/Firebase may want on-device capabilities limited in order push adoption of their cloud ML services.
The WWDC 2018 ML sessions showed some solutions to address these issues. (I don't exactly recall which as I'm not using ML yet, so I skimmed through them.)
Footnote: You can do a "local" login on a Linksys smart router, but if your internet is not working during setup it will refuse to even initialize. It made me reset the NVRAM because my internet wasn't connected yet. If you do refuse to create an online account, the internet service is still enabled by default.