When submitting to (computer science) conferences, you usually have full control over the layout and are able to submit a camera-ready version that will be published as submitted.
For journals, however, you usually have to submit a manuscript that will be reformatted before publication. In my experience, this includes rearranging and scaling of the figures, reformatting the equations etc.
Of course, it made sense before researchers were able to typeset camera-ready papers, but now it seems to be unnecessary and just generates additional costs and errors. Why do publishers still prefer this way instead of letting the authors do the layout? I would expect that the journal style can easily be enforced by providing an appropriate LaTeX template and some additional rules.
I thought it might be due to copyright issues (maybe if the publisher does not do any modifications, it has no right to have a copyright on it), but it also happens for open-access journals where this should be no issue.