I don't think this is exactly correct, but I think it will work better than what's there.

((string-match "\\([+-]?\\([0-9]+\\|[0-9]*\\.[0-9]+\\)\\)e[mx]" spec)
    ;; Character based
    ;; XXX: should be relative to font size of enclosing element
    (string-to-number (match-string 1 spec)))

Basically, each "em" unit should be the size of two spaces (or lines).  Currently the function uses font-spatial-to-canonical, which returns the number of points for the number of pixels.  Somehow, this number of points gets turned into the number of spaces, which is definitely wrong.

If you can think of a better way to address the issue, this was just an idea.  It was pretty much just a hack to make a page not look totally ridiculous, just thought it would help the project.

Colin