When the CPU's developer manual is 2198 pages and still growing, any option other than starting with Linux and trying to keep up (i.e. VT, SGX, TPM, GPU, etc) will be extremely costly.
I think that Intel’s cpu complexity (and arm and for that matter RISC v CPUs) is that they are designed to speed up existing software. Software that was written 50 years ago on simpler machines. Starting from scratch is not as complex as you might think if running existing software at “native” speed isn’t a requirement.
Are we going to give up GPUs, virtualization extensions, SGX enclaves, memory barriers, upcoming AI-acceleration, etc? To my knowledge, there is no software concept that will performantly render these functional units obsolete.
If we keep the chip features, the OS will have to wrap them in one way or another.
edit: my personal preference would be to go full-Terry, but I know that's a pipe dream.