X-headers are very useful for experimental features and private agreements -
in fact, x-headers have been quite a problem. they are excellent in
theory, but the header can become popular, thereby becoming a de facto
standard. then we are stuck with a defacto standard that uses and x- label.
I don't know where you get the idea that x-headers have been quite a
problem. As far as I can tell, they've been quite successful.
Offhand I cannot think of a single read/write X- field that has become a
de facto standard, but I can think of several X- fields that are really
bad ideas and which should never have been deployed even in a single
implementation. But at least they're easily distinguished from other fields.
OTOH, most of the poorly designed fields that have gained wide deployment
don't start with X-.
To me this says we should strengthen the X- convention rather than abandon it.
simply registering headers does not encourage or discourage private or
public efforts.
What it does is to encourage people to use ordinary-looking field names
for poorly designed fields which don't have the benefit of public review.
To the extent that this has already occurred, it's almost universally been
harmful.
Keith