The bottleneck when I read is not the time it takes to recognize the words, but the time to understand the ideas and relate them to my experience. It can cost me an hour to read a page of a computer science research paper involving enough mathematics.
In my experience, reading papers with math gets a lot faster as you get used to the conventions. It used to take me almost a whole day to digest a single math or CS paper in a field I was familiar with, but now I can get through a couple in an afternoon, provided that I am not interrupted.
The "chunking" that you develop is like that in chess, or programming for that matter. This seems to be one of the primary reasons why journals will reject papers with unconventional notation. Unfortunately, the notation doesn't seem to stay uniform across disciplines (math <-> physics) or even languages (english <-> french) even when using the same mathematical structures.
