Personally I use mostly labels, since they make looking for citations quick and easy by letting you search by multiple criteria. Looking for citations on the transpiration of urban trees? Search for "tree physiology" and "urban trees" labels. Want to narrow down the list to experimental studies? Add "experiment" label to the search criteria.
I believe labels also allow more flexibility - with folders, you need to decide early on what structure to use, and stick to it. In my case, I couldn't think of a clear way to organize papers in my database using folders, as most papers are interdisciplinary or would thus belong in various folders. Using labels eliminates this problem.
The only use I have for folders at the moment is to gather citations for a specific paper, or to mark papers that I'm currently reading or planning to read.
My example: I work on urban climate and use around 30 labels. They are grouped (and formatted) by:
- climate zone (e.g. temperate climate, Mediterranean climate) - orange labels
- methodology (e.g. review, experiment, observation, GIS, modelling) - blue labels
- subject (e.g. tree physiology, urban heat island, thermal comfort) - red labels
- optional: focus on a specific part of green infrastructure (urban trees, urban parks, green roofs). - green labels
This way I can easily see if all necessary labels were added to a paper, but most of all, I can quickly search among my papers using multiple criteria. In fact, this was exactly the reason I migrated from EndNote to Paperpile