When the adaptive mode is off (with the default DSubPix setting), the map generation will be consistently fast, and only dependent upon the resolution of the high-res mesh. When the adaptive mode is on, the amount of time will vary and will mostly depend on:
The amount of distortion applied to the mesh (less distortion=faster while more distortion=slower).
Whether you are using the existing low-res cage or importing/morphing to the pre-edit state (which usually requires more calculation time).
More detailed information is available in this thread.