I've had two different conversations in the last couple of days about how people choose where to submit papers, and it's a decent topic for a blog post. I can only speak for myself, but I think I'm pretty typical. To frame the discussion, consider why we publish journal articles in the first place. We want the scientific community to know what we've been doing, so that our work can be built upon - if we've answered a question that many people want answered, those people should know. If we've developed a new technique that will be useful, or if we've learned something that changes the way we think about some (ideally important) system, the rest of the community should know. Of course, publications and citations are also one metric of performance. It's a marketplace of ideas out there, and if no one cites your papers, then that says that you may not be having a major influence in moving the field forward.
The desire to disseminate knowledge and get recognition both provide a motive to try to publish in the highest impact journals that are appropriate. On the other hand, not every publication-worthy result is necessarily earth-shaking in significance. I know that there are some people who apparently send every halfway-decent paper to Science and Nature first, because "why not?" I tend to be more conservative and self-assessing. Not everything is of interest to a broad readership. Similarly, there are physicists who send every result to PRL. Again, let's be honest - not every physics result is PRL-worthy. Furthermore, in the nano arena, sometimes the chemistry or engineering literature may really be more appropriate than Phys Rev, and that's fine. I do try to aim for the highest "impact factor" journal that seems topical and reasonable - that's just common sense.
An additional factor is the time-to-publication. If you're working in a competitive area, you may want to get a result out in the peer-reviewed literature fast, and the best way to do that may be to publish in something other than PRL. The arxiv mitigates this a bit, but not all publishers like electronic preprints.