A content provider (such as Amazon) who wishes to annotate users pages should be able to implement it once, using open standards, and have it work in Google, MSN, or Yahoo. Users should be able to chose whichever content providers they want, and it should be totally open.
So Joshua mentions users and content providers, but not authors. There's always that tension between authors having ultimate control, and users having ultimate control. There can be only one.
Even if you tried to use markup to indicate which sections of a document were 'annotatable', end users will still want non-marked up documents to have annotations. Identifying a region of a display to place annotations might work for some uses, but in-place links and coloration are extremely user friendly much more so than a parking lot in a side banner.
And what about non-HTML document formats? Sure, Microsoft has MS-Office Smart Tags, but who's doing the RSS (Really Super Smart) tags? What about SMIL? Or media playlists? Should the Web stay within the bounds of what transclusion supports? Or should it go beyond that to enable client applications to transform documents outside the control of the author? And what will be possible when non-HTML client applications start doing this?