> Handling data in/from a filesystem seems to be better
> understood/easier than via some new interface.

I don't object to doing kernel configuration through a filesystem, but
I do object to the misconception that a filesystem interface isn't
actually a "new interface."  Even if it's as simple as reading or
writing an ASCII decimal number from a file, magic filesystem
interfaces are new interfaces and need to be documented just like
new system calls.

> Plus it follows the traditional Unix "everything is a file"
> philosophy.

Not really.  The Unix philosophy is that most shared data objects are
files (although some objects are not, like unnamed pipes).  Using file
data as a control channel is a relatively new innovation and has a
very different feel to it.  NetBSD has traditionally resisted this
innovation, perhaps for good reason.