This is not to say that paragraphs should be mixed and matched at random, wholly and irretrievably separated from their original context. The user should always have the option of reading an authored work in its entirety, and in linear order if the author conceived it as a unilinear work. But for the purposes of creating a quickly navigated, flexible information space, units must be small enough to be apprehended quickly, almost at a glance.
Admittedly, the paragraph is too small, in most cases, to sensibly stand alone.
The danger in stringing diverse paragraphs together is that you lose the transitions. In a fragmented hypertext environment, when users branch off onto a new path, they must be able to see immediately why a link exists between where they were and where they are. If they are dissatisfied with this new path, then they must be able to return to the previous place quickly and easily.
Since it is necessary to take both speed of movement and immediate comprehension into account, in most cases a document will have to be cut down into self-contained subsections, probably consisting of several paragraphs.