The buying practices necessitated in order to keep a large bookstore financially viable these days have skewed the kinds of books that are deemed saleable profoundly; the redemptive promise of the Web was that the magical long tail might create markets for even those niche publications that have been edged out of mainstream publishing and book sales.
And yet (as I understand it - corrections welcome) for a book to be sold in more than one place online it must be equipped with a set of tags (ISBN, summary, thumbnail image etc) according to a metadata standard. Without these, the multiplicity of bookselling affiliate schemes, APIs and so on will not be able to carry the title, and the book will not sell. And this additional informational labor is beyond the technical and time resources of many small publishers. So while a bookstore (in its ideal, pre-Scott Pack form at least) might be imagined to carry a genuinely serendipitous mix of local publications, the manager's choices, remainders, bestsellers and second-hand titles, this slick performance of serendipity relies on several intricate but invisible additional layers of technologization. Thus, while it gives the feeling of serendipity, the data architectures required to sustain the 'bookstore' metaphor push the available selection ever more towards a literary monoculture.