A successful compromise might still be a denial-of-service attack, but by immediately taking the compromised system offline you can have confidence that your computers aren't really "owned" by a bunch of hackers.
> Iff linux and much of the userspace were redesigned you _might_ be able to use it to detect rootkits, but even then its unlikely to help... Attackers already aren't rebooting your systems into new kernels for rootkit purposes: They usually use intentionally exposed features (or bugs) to add code to the kernel without rebooting. ... so TPM would attest to you that it booted your trusted kernel but it wouldn't matter.
The idea is that the signing process continues after boot and no code is run without first being checked. If you can extend that to all code in the system (which I agree is a lot of work) then you can detect code changes after boot. Presumably NX bits help here too.