It's not clear to me how a reader would even be able to distinguish which version they were reading (and wanted to improve.)
I can't conceive of orchestrating simultaneous corrections of one error to multiple custom targets where the linkage (shared content) among them isn't explicitly identified.
If a desired format isn't available for a given device or style (probably what is called in CMS-speak as a theme) then the work needs to be directed to improving the generalization of the master format and the exporting software that converts that master format to the target format.
One warning about HTML. Remember that its heritage is coming from SGML, whose intent was to encode all syntax and display information, to remove the syntax part and provide just the display part. So it's not accidental that we end up needing to implant syntax.
Also remember that HTML requires "well-formedness", and that books are often not syntactically "well-formed." in that sense. (One reason XHTML is losing favor. It's designed for software developers, not creative writers whose organizational instincts are more fluid.