I think the term linear is really inappropriate in this discussion. Linear
is indeed adopted from mathematics, where it referes to a function which
creates a straight line (i.e. with no curves or corners). To give an example
suitable to the current context, reading-speed could be called linear when a
reader reads a book at a constant speed of 1 minute per page.
When hypertext is discussed, sequential is used more often. However, that
doesn't add all that much to the discussion - very few people (split-brains
forming the only known exception) can read more than one stream of text at
the same time. Whether this stream of text is in a book or on a web page
doesn't matter all that much.
So what's the discussion about then? Probably the best distinction to make
is between fixed sequentiality and non-fixed sequentiality. My mail is a
good example for the first - it's very hard to read this mail in any other
way than from first to last, as I've not written paragraphs that can be read
and understood by themselves, as each paragraph specifically refers to
information presented in the previous paragraph.
In other words, it's a structural thing. Although hypertext is easier to
navigate, a bookcase encyclopedia is structured the same way as a hypertext
version, and, more importantly, can be read in exactly the same way. The
same goes for many other reference and scientific works.