What you're getting at is important, and described really well in a recent article by Judea Pearl on Simpson's paradox [here](http://ftp.cs.ucla.edu/pub/stat_ser/r414.pdf). Namely, because they didn't manipulate words spoken directly the evidence for more words causing an increase in performance is conditional on your/their beliefs about the structure of the world.

I would say that many scientists are good at acknowledging this, though they may get over eager and forget to emphasize the point once they get to their discussion (or when an article is summarizing a piece of research).

Whenever I read observational research, I just mentally insert "under a set of structural assumptions about the world" before every statement like the one you listed above, and everything is okay again :).