Nicole Henning suggests that academic libraries and scholarly presses work together to create the ultimate mobile app for scholarly ereading. I think about the requirements a bit differently, in terms of the functional requirements.
The main functions are obtaining materials, reading them, organizing them, keeping them, and sharing them.
For obtaining materials, the key new requirement is to simplify authentication: handle campus authentication systems and personal subscriptions. Multiple credentialed identities should be supported. A secondary consideration is that RSS feeds (e.g. for journal tables of contents) should be supported.
For reading materials, the key requirement is to support multiple formats in the same application. I don’t know of a web app or mobile app that supports PDF, EPUB, and HTML. Reading interfaces matter: look to Stanza and Ibis Reader for best-in-class examples.
For organizing materials, the key is synergy between the user’s data and existing data. Allow tags, folders, and multiple collections. But also leverage existing publisher and library metadata. Keep it flexible, allowing the user to modify metadata for personal use (e.g. for consistency or personal terminology) and to optionally submit corrections.
For keeping materials, import, export, and sync content from the user’s chosen cloud-based storage and WebDAV servers. No other device (e.g. laptop or desktop) should be needed.
For sharing materials, support lightweight micropublishing on social networks and email; networks should be extensible and user-customizable. Sync to or integrate with citation managers and social cataloging/reading list management systems.
Regardless of the ultimate system, I’d stress that device independence is important, meaning that an HTML5 website would probably the place to start: look to Ibis Reader as a model.
The authentication and DRM is the hard part, which is not entirely under the control of developers. There are certain auth and DRM requirements insisted upon by platforms and content providers which do not have good solutions in the software you imagine.