Why not port Linux kernel to Common Lisp?

Conventional wisdom states that OS kernels must be written in C in order to achieve the necessary levels of performance. This has been the justification for not using more expressive high level languages.

However, for a few years now implementations of Common Lisp such as SBCL have proven to be just as performant as C. What then are the arguments against redoing the kernel in a powerfully expressive language, namely Common Lisp?

I don't think anyone (at least anyone who knows what they are talking about) could argue against the fact that the benefits in transparency and readability would be tremendous, not to mention all the things that can't be done in C that can be done in Lisp, but there may be implementation details that would make this a bad idea.

A DSL for memory and drivers? Why not write them in good suitable, already existent language to start with?

A layered approach is a better idea in my opinion. Write from the metal towards an abstract machine specification, which includes things like memory mapping and address spaces, processes, file system and drivers. Some things in this layer could be implemented in lisp if proper hooks are provided.
This then plugs into a lisp implementation which constitutes the upper part of the kernel and the user space.

Instead of forcing lisp to fit a machine, make the machine fit lisp