Well, I use both. Deconvolution as step 1 after integration, and UnsharpMask as one of the final steps.
Unless you want to use the data for scientific purposes, and lets face it, most of us are amateurs, I think the main target is to make an image with as much visible information as possible. That means that all tricks are acceptable, as long as it "looks good".
Deconvolution can also produce "panda eyes" around stars, but with both processes using a star mask and finely tuned deringing can solve that. Ringing in itself isn't a bad thing. When you resample using a Lanczos algorithm it uses a slight ringing, and that can produce excellent results when upscaling.
Using the PSF is a great idea, but often you are able to get equally good (and sometimes better) results, by tweaking the manual settings.
Best regards
Morten
Edit: Even though Deconvolution works best on linear images, it's also a fantastic tool when working on non linear images. I do that a lot. Actually, the deringing sliders are easier to use on nonlinear images.
Oh yes, I also forgot: Deconvolution also gives you the possibility to reduce noise slightly, whereas UnsharpMask is pure ringing, but you can get an almost similar effect by using a luminans mask together with sharpening.