I find it a mayor drawback of Google Scholar that it also finds things which are not publications at all (e.g. sometimes it finds talks which end with a bibliography page).
Another drawback of GS is that it gets confused by some preprint-series I know, namely the ones which attach the list of all the preprints in that series. In these cases GS counts these as citations which is just wrong.
However, I find GS helpful and use it frequently but I would not consider its output reliable. (Another related issue is, that GS sometimes identifies weired things as the author, e.g. try to find papers with "Mathematik" as author...)
By the way: Wouldn't it be a great feature of both GS and MathSciNet to give snippets of the places in which a paper is cited? Imagine, if you click on the "cited by"-button and see a list of paragraphs or sentences is which the citation happens...
