Each of your 'examples' has errors of omission. Really serious ones. You refuse to show areas of smooth tone and critical grain detail - why? And then when challenged, give an 8x10 - which has the exact problems I outlined. If you'd aimed the scan for 1000-1200ppi or so, it might have looked pretty good. Furthermore, why does the Hasselblad scanner (in the non-downsampled examples I gave) quite clearly deliver reasonably accurate depictions of HP5's granularity all the way from 1500ppi up, yet the Epson doesn't?
And if you really knew anything about convolution problems, you wouldn't be using the most basic USM to try and resolve them - deconvolution sharpening and HIRALOAM in one form or another are often less obvious in their side-effects and more effective. The Epson still fails because of its fundamental shortcomings in resolution of granularity and small fine details in those lower contrast areas. But that would require you to accept and understand that MTF and noise have far more impact on overall resolution than any high contrast resolution target does.