I feel like there is a lot of misunderstanding here. As I RE'd a lot of TX's code and delved deep into the workings of EmuNAND I can tell you how it works.
First misconception: The EmuNAND on internal NAND does not swap entries in the GPT on the fly. It simply redirects all eMMC SDMMC IO read/write blocks by changing the block address. The EmuNAND on microSD added support for enabling microSD early in the system boot process, but from there it's also just simply redirecting IO to microSD and changing the block address too.
Second misconception: Did you actually look at the thermosphere code on github? I'll point you to it:
Where's this so-called SDMMC driver for Thermosphere? There's nothing there!
And besides, there are so many possible SDMMC drivers out there as the Tegra X1 is a bog standard SD Host Controller interface. So it doesn't need to be a driver specifically written for the Switch.
At this point the Hypervisor method is not going to bring much more 'proper' to the table. In fact I'd wager it would make the system much much slower, because that Hypervisor context swapping does not come for free