It does indeed vary from subject to subject, and journal to journal. I once got in a short argument with some math students who had asserted "If its going to be published, it needs to be in LaTeX", a disagreement that only ended when I went and found some submission guidelines. For three fairly good journals in my field (Epidemiology), you have some considerable differences.
American Journal of Epidemiology wants everything in either Word or PDF format - LaTeX documents are compatible with this, but its certainly not doing anyone any favors in terms of already being formatted. Epidemiology will accept LaTeX documents, but warns that the odds of typesetting and other erros increases in formats besides Word. Journal of Clinical Epidemiology only requires that an editable version be available.
So the de-facto standard in my field is Word, though of course there are ways around that. And then, as mentioned, there are formatting issues beyond just "what's the file extension?" How references, the text, and figures are reported - must odds ratios be graphed on a log scale or not?. How p-values are treated. What format graphics are allowed to be in, etc. As Fabian said, submitting to a new journal often involves combing through the same content to subtly tweak formatting.