> UEFI let’s the FW signal to the OS upon boot “please install this driver blindly”, but it can’t force the OS.
> Only Windows obliges to this request. Linux is immune.
UEFI firmware runs before the OS and with greater privileges – e.g. it has control over System Management Mode (SMM), which is hidden from the OS. It can force the OS to do whatever it wants. For example, it can modify system files on disk, which should be enough to compromise most Linux installations. If that isn't possible (because something verifies the files), it could, e.g., patch the next stage bootloader it loads (which could be GRUB or the kernel itself via EFISTUB), in memory before executing it. Or, as an easier approach, it could just add a SMI handler that patches the kernel later on.
It is true that Windows 8 and later exposes a way for UEFI to "ask it nicely" to load a given driver, without having to patch anything, namely the WPBT (Windows Platform Binary Table). As you note, this is the mechanism Lenovo used to persist Superfish. The reason it exists, however, is for anti-theft software such as Computrace/LoJack. Earlier versions of Computrace took a "brute force" route to installing its persistence driver, patching system files on disk, and there wasn't really anything Microsoft could do to stop it. So instead they decided to give them a sanctioned route to accomplish the same thing, which at least is less likely to break something in the process.
However, at least according to the white paper, the LoJax UEFI rootkit does not use WPBT at all. Instead it uses an approach that seems to be inspired by older (pre-WPBT) versions of Computrace. [1] Specifically, it drops a binary "autoche.exe" into the filesystem, then modifies the registry to execute autoche.exe on boot instead of the normal autochk.exe (note that the last letter is different).
Thus, WPBT is irrelevant in this case. The malware hijacks Windows by modifying its filesystem, and if the authors cared enough to target Linux, they could hijack most Linux installations the same way (and all installations with a somewhat more difficult approach). But then, even if some piece of malware did decide to use WPBT when targeting Windows, it could still use the more elaborate methods to target Linux. At most, WPBT makes it slightly easier for UEFI malware to target Windows than it would otherwise be. But it really doesn't make much difference.
[1] To clarify: As the white paper describes, a later stage of the bootstrap, rpcnetp.exe, is actually copied from an older version of Computrace and merely patched to change the C&C server address. However, the UEFI part of the malware is custom, and merely uses a similar (but slightly different) overall approach. Computrace's own UEFI driver worked by actually modifying autochk.exe, whereas the malware modifies the registry to execute a different filename instead.