Somewhere between the original (image 1) and deconvolution (image 2). Image 3 looks too heavily processed - it may have work have worked if you have more data to play. You also need to be careful as to what stage you apply deconv or unsharp masking in your image processing scheme. If you lose critical information too early, then no amount of processing will bring it back. It is not very common to sharpen an entire image, simply select areas you wish to bring out. This will minimise data loss and provide a natural feel.
Its a matter of personal opinion. There is no right or wrong.
