At Imazen, all of our own websites are created with a mix of 95% Markdown (Kramdown) and 5% Slim.
The trouble with any XML-based solution is that it's resistant to versioning, diff generation, and automated merges. Only context-free syntaxes (such as Markdown, Wiki/Creole, etc) can effectively be used in a distributed version control system (like git) or an offline/organic workflow.
We started the Repositext Initiative to offer publishers and large-scale content managers a better solution. Our first publisher to use the system translates content into 70 languages and has several thousand books which are translated from English to other languages. Many translators on the African continent lack internet connections in their city, and have to mail USB drives or commute to an internet cafe in a different town.
The master version of all content is stored in Markdown form (the MultiMarkdown/Kramdown variant, in order to support named paragraph and character styles).
Using context-free tokens allows us to preserve additional metadata attached to the text (such as indexing hints, translation segmentation marks, and paragraph/record identifiers). These can be automatically merged as content changes, using the Suspension library
Enabling repeated import from Word, InDesign, and other formats enables the system to support any type of workflow that happens - no linear process is enforced, and users can use the tools they prefer.
Our feedback from authors on the Markdown syntax has been extremely positive - not only for the simplicity and readability, but for the ability to see all data at once, and to detect formatting changes visually within diffs.
On a large scale, Markdown is even better than on a small scale.