I definitely think adding COinS to OpenLibrary pages could make sense.
I'm curious what everyone's use case is. Is it mainly browser plugins
that can inject links to a relevant OpenURL router so that you can
find books in your local context? If so I think use of COinS in
OpenLibrary makes a lot of sense.
There is an orthogonal use case of making structured metadata
available via a book display. I'd personally prefer to see web pages
for books include auto-discovery links for alternate machine readable
representations. This is how blogs are typically tied to their atom
and/or rss syndication feeds.
For example if the web page for Weaving the Web  could include
<link rel="alternate" type="application/json"
<link rel="alternate" type="application/rdf+xml"
The JSON one works now of course, the RDF one is hypothetical. A
consuming application like Zotero (or a web crawler) could then use
simple auto-discovery to find the machine readable data. Another
alternative would be to use RDFa to interleave metadata into the HTML
display itself. We have nice things like the Bibliographic Ontology,
and the emerging RDA vocabulary that you are working on which would
fold right into these RDF representations. This machine readable
metadata could also link to things like the PDF, table-of-contents,
etc where appropriate.
It would be great if Brewster's idea of "a web-page for every book"
could also mean machine readable metadata for every book. OpenLibrary
has a rich database available behind it it, and it seems a shame not
to expose it in the HTML in a web-friendly way.