> It may even be time for dedicated kernel cpus/cores.

Oh yes, I agree! One needs to be able to phycically (un)lock the "kernel fpga" like a door without remote capabilities, except for server cpu's. Or whatever chip designers believe is a good "physical kernel embodiment" other than fpga.

EDIT: I know it's not really clever, but I would really enjoy hearing any solutions that doesn't try to fix it at the hardware level.