The Nvidia DDS plugin for Photoshop has some great controls for making the MIPs. I like using the Mitchell MIP Map filter, tends to do the best job.
For sharpening I use the Unsharpen Mask filter in the DDS pplugin. First I make the DDS without any sharpening, the close and re-open it, choosing to display the MIPs. I then use Photoshop's Unsharp Mask tool to find out what settings work best, then use those in the DDS tool.
You can learn more about filtering here...
As for sizes, for us it's all about how close we get to the model (what kind of camera behavior we have), and how we juggle our memory. There are still going to be limits and bandwidth problems.