Open any random sample of science journals, and you will find a range of styles of scientific paper, from brief 'letters' or reports occupying only a few pages to monographs many tens of pages long, and from terse accounts of new results to opinionated commentaries and weighty reviews of broader fields.
The conventional wisdom is that research papers and review articles require different editorial styles, reflecting their different goals and audiences. From my own experience as an editor on two rather different types of publication, I would certainly concur. The remit of a research paper, the bread and butter of my work at Science  , is precise: to document the authors' investigation. For the vast majority of papers in the scientific literature as a whole (if not in Science) the audience is quite small and tightly focussed: Indeed, many are read from start to finish by no more than 10 people--that's including the authors' colleagues and the referees.
The style of a research paper is quite formulaic, with most journals following some version of the standard Introduction-Methods-Results-Discussion sequence. With the help of referees, the editor's job is to ensure that the study holds water and is a genuine and novel advance, that it is presented clearly and economically, and that it is placed in the wider context of relevant developments (historical and contemporary). Most importantly, the editor needs to be satisfied that the paper contains enough information to enable the reader to judge the merits of the work and--in theory--to repeat it. Of course, some work is not repeatable--a long-term ecological data set, for instance, is effectively a unique resource, and in such cases the editor needs to ensure that the data are sufficient to allow readers to assess the validity of the authors' interpretation of the results.
Before joining Science 3 years ago, I worked for Trends in Ecology & Evolution  (or TREE, as it became known), a journal devoted entirely to news, comment, and reviews. A review paper generally has a more diffuse remit than an original research paper: to document and interpret the developments and state of the art in a given field. The audience for a review is larger than that for the average research paper and will typically span a range of familiarity with the field--from the beginner or interested outsider to the specialist. The style of a review is less formulaic, and editors have lively arguments, among themselves as well as with authors, as to what constitutes legitimate review material. Should a review concentrate strictly on reporting the advances made over the past few years, or can it be extended to include new syntheses, new hypotheses, or new conclusions? The only point of consensus seems to be that a review is defined by what it doesn't contain: namely, original research and new results.
A key contrast between reviews and research papers is that reviews tend to be commissioned (i.e., invited) by the editor rather than submitted like research papers. Topflight research journals have high rejection rates for research papers--sometimes exceeding 90%. Under these circumstances, a lot of competent papers have to be turned away after peer review, even if the flaws identified by the referees are mostly fixable. By contrast, the rejection rate for invited reviews is far lower. This is partly because the editor commissions with an eye to filling a set quota of pages but also because a key element of the quality control has taken place at the outset of the process: The editor has tried to choose an interesting topic and an author who can be expected to deliver an authoritative article. Review manuscripts will, nonetheless, often receive criticism from referees, usually on the grounds of balance of coverage--a flaw that can usually be fixed by the author. Often, there are simply differences of opinion, which--because review papers are almost bound to contain some subjective elements--do not necessarily lead down the road to rejection.
Despite these differences, the editor has to bring the same basic skills to both types of manuscript: an eye for detail, a nose for what is and isn't interesting, and a feel for helping authors get their message across more effectively to the reader. The editor also needs to be prepared for flak from rejected authors, although this applies more to research papers than to reviews, partly because it happens more often, and partly because there is often so much riding on the success of a research paper. Because scientific careers are partly built on the publication of original research, it's not surprising that rejected authors can get upset. Hence another very necessary set of editorial skills is to appreciate the responsibility of the job, to build confidence in one's own editorial judgement, and to communicate decisions effectively and sensitively.
Once a paper, of whatever type, has been approved for publication, several challenges remain for the editor. One of the most universal is to cut the paper down to the desired length. Most journals stipulate an upper word limit, and a maximum number and size of figures, tables, etc. I have had the good fortune (or ill luck, depending how you look at it) to work for two journals where the length limits are particularly tight: Reviews in TREE are supposed to be 2500 words, with 35 references; Reports in Science are also 2500 words--but that includes references and legends, too. In my experience, most authors routinely ignore these requirements and either assume that an extra 25% will go unnoticed or protest to the editor that their work merits exceptional treatment. Neither of these is usually true, and the editor can generally find ways of pruning without losing essential information. Although the editor should always strive to preserve the author's style, a lot can be achieved by editing out unnecessary verbiage or by using shorter words. Thus, 'resources required for metabolism and continued growth'--one of my favourites--becomes simply 'food', with no loss of information.
Length limits exist for two reasons. First, there is the economic argument: Longer papers mean more pages, which mean longer editing times and greater production and distribution costs. Second, and just as important, the needs of the reader must be considered: On the whole, shorter papers tend to be read by more people, and if the goal is to reach an interdisciplinary audience, a tight length limit is always desirable.
And there are other ways of streamlining a paper. In the past few years, the possibility of placing extra information as an online adjunct to a printed paper has made this process easier and has made a journal like Science more amenable to papers in areas that were often off-limits because of the sheer quantity of data required to support the authors' claims. There is a downside to this, of course: The editor still has to check that the printed and online information is correctly balanced, and that--released from any length restriction-the online material itself is not too daunting or unappetising. But this is undoubtedly the direction in which scientific publications are heading.
What are the rewards for the editor? First, there is the very tangible reward of producing something new every week. Second, you don't have the anxiety of applying for research grants. But the best part for me, both at Science and at TREE, has been the privileged access to a wide range of new and fascinating research, and the opportunity to interact with a huge number of scientists across a range of disciplines.