That kind of thing makes indexing a nightmare but it's not any more difficult to archive than .exe or .swf files. Sure you need a "player" but that's true of PDF and OOXML and people don't really complain about those.
In a PDF or normal program there are clear semantics what is text and what is something else.
On a canvas, everything is just made up of pixels. You'd need OCR Software to detect what is what and they won't ever be 100% correct unless you use only text and fonts which are made to be recognized by OCR Software.
Ive implemented a text editor with screen reader support using the html canvas element. Every graphics interface is just a canvas, what is sent to the screen is just data. The nice thing with html is that the data is human readable.
