While griping about the hassles of typesetting math in this Typophile thread, I finally put my finger on what makes so many otherwise good typefaces fall apart in math or technical work: character-level legibility. A good text face works best when its letters work together to make good word shapes, right? When the individual glyphs don’t pull the reader outside of the flow of the text with too many quirks or spacing irregularities. The trouble with setting math or other technical material (chemical equations, charts of ID codes, etc.) is that the context for the individual letters is much less familiar than in typical text. If text is comprehended word by word with less need for the letters themselves to be individually distinguished, then math is read letter by letter in such a way that almost any character could be swapped out for another and change the meaning entirely.
Most of my problems setting math over the years have had to do with letters that just aren’t unique enough when you pull them outside of normal text and start mixing and matching them with Greek and symbols and numbers and lord knows what else. Especially once superiors and inferiors are used, it becomes absolutely critical to know if a glyph is an “l” or an “I” or a “1” or a vertical bar, for instance. (If you’re seeing the right CSS styles for this page, see how nicely Georgia distinguishes those from one another? Check out Arial: l I 1 | )
A good face for this environment needs to strike a balance between the ability of the letters to combine easily for typical reading comprehension, but still hang onto enough unique appearance to hold their own in the free-for-all world of tables and equations.