I don't remember whose research it is (it's a bit removed from my field of expertise =), but the gist of it is that they tried to highlight different parts of speech in running text in different colours, and showed that it actually decreased reading speed and comprehension.
Of course, there are several possible error sources (unfamiliarity with the colouring scheme for example), and whether this can be transferred to reading code is an open question. But I still think it's important to keep the existence of this kind of research in mind when discussing this kind of question (and also the similar question of mouse-driven interfaces vs. keyboard shortcuts), as it easily devolves into strongly held opinions and arguments along the lines of "it stands to reason that $opinion-held-is-true"
Of course, there are several possible error sources (unfamiliarity with the colouring scheme for example), and whether this can be transferred to reading code is an open question. But I still think it's important to keep the existence of this kind of research in mind when discussing this kind of question (and also the similar question of mouse-driven interfaces vs. keyboard shortcuts), as it easily devolves into strongly held opinions and arguments along the lines of "it stands to reason that $opinion-held-is-true"