AFAIK there’s no significant difference in how current Blender and OpenSubdiv handle creases. The difference with OpenSubdiv is that working at 5 subdivisions is actually viable.
Also, you can just use the Mark Sharp functionality, it works fine with subsurfs.
I tried to recreate your test in 3ds Max using the OpenSubDiv modifier at 3 levels of subdivision and only fully creasing edges. The strength of OpenSubDiv is that it will display the creasing in Blender exactly the same way it will in Max so it is a fair test in my opinion. This is the results.
I see in Max you see the same shading problem - and thanks for the mark sharp tip - note to myself: why did I never think about that.
I am curious now if in TS they do the same full crease with mark sharp because you can get even with not very sharp corners prefect and sharp creases. In my example the corners are more like 90 D.
Too bad I was hoping Pixar’s subdiv might deliver something like that.
I’ll throw here the big question i have about OSD: since we’ll have to set it as the last modifier in the stack, how can we use the increased mesh-resolution in terms of displacement? A further displacement modifier would broke OSD efficiency. So, is the experimental subdivision/true displacement about to get fully developed? Or is it incompatible with OSD anyway?
I imagine at some point displacement will either get rolled into the Subsurf modifier, or Sergey will program in a special case for where displacement is after Subsurf.
Indeed, I was hoping that Pixar could solve the creasing in a good fashion but it is not a miracle cure unfortunately. If I had to guess it is just an extension to the catmull-clark subdivision algorithm and as such it falls short on the same pitfalls that the regular subsurf does (uneven polygon distribution gives shading errors and so on)
However I did notice that it does handle creasing a bit better than Blender does currently.
fully developing what is in expermenatal state now would rock: node based displacements on top of adaptive tessellation.
I believe Arnold works in a similar way (user-side)
<b>Subdivision and displacement</b>
Arnold supports Catmull-Clark subdivision surfaces. Subdivided vertices are then vector-displaced through arbitrary shader networks. High frequencies can be automatically captured as bump map, reducing the need for excessive subdivision.