Since MediaWiki allows images that are clearly separated out, we should see if we can use <figure> and <figcaption>. These tags are intended for "some flow content, optionally with a caption, that is self-contained and is typically referenced as a single unit from the main flow of the document."
This would particularly make sense for thumb and frame where a caption is already shown. However, figure can also be used without figcaption, which should simplify implementation.
The main concern is backwards-compatibility with browsers that don't know about these tags. They're new to HTML5 (which is now always on in MediaWiki), but
- Older IE will not be able to style these elements from CSS
For issue number 2, I think jQuery's internal createSafeFragment already takes care of old IE. So only the styling issue remains.