It isn't a question of blame, it is a question of whether or not
agents have been generally able to construct HTML objects that
are legible without any formatting. In hindsight, the answer to
this question has proved to be "no". And since text/html's
utility was predicated on the answer being
"yes", it was a mistake to have defined it.
I disagree with the logic here: it's like saying that the creation
of weapons is predicated on their being put to good use, but having
seen them kill people, the creation is a mistake. At the end of the
day, it is the responsibility of the agents to label the content
correctly... you can't hold the tools creators responsible for
their misuse.
If the agents consistently abuse the protocols and standards, then
there is an argument that the agents needs aren't being met. HTML
is abused because people needed the equivalent of RTF, but wanted
to piggyback on the (supposed) interoperability/business benefits
of WWW standards.
Likewise, agents consistently send data as text/html when in
reality, the content is application/x-whatever. The main reason is
because again, of (supposed) interoperability, and because they
didn't have any escape route.
With HTML, the genie is already out of the bottle, but with XML
there is a chance to get MUA's working as they should: when
sending textual data, use text/xml, but when sending application
specific data, send application/foo+xml, etc.