In general, shorter windows contain less information, and will thus provide less frequency resolution. But short windows will be more localized in time, and thus a sequence of short windows might allow greater resolution in locating or separating time domain behavior.
Longer windows will contain more information, but this information will be averaged across the entire window duration, thus potentially smearing out any transients or other non-stationary behavior that might change more rapidly than the window length or duration. But the greater information content might allow a higher resolution (both for separating peaks and locating or estimating isolated stationary frequency peaks).
Overlapped windows will provide some partially redundant information, but perhaps better localized data regarding rate of phase change for some types of signals that are stationary across multiple frames.
Added: Overlapping is also common when a non-rectangular window is used. Since the windowing process (combined with finite arithmetic) can be quite lossy at the edges, overlapping helps provide some of the information that was thrown away by the windowing.