One of the things not mentioned yet: scripting!
Incorporating Python into Poser was arguably the best move ever made in the history of the app. Smith Micro can't predict everything users want, there have always been those weirdos that use Poser in unusual ways, but if everything that Poser can do is also accessible via well-documented Python calls, there will always be a Python scripter that'll build the desired functionality.
So what I want most in the next version of Poser is that EVERYTHING is scriptable via Python.
What would be even better: Poser as a Python library, so that it can be automated from the outside. Then it would be possible to script Poser into an automated 3D workflow.

I'm not really interested in content provided with Poser itself. I buy the content I like at CP (animals), Renderosity (clothing and hair), DAZ (human figures, environments) and RDNA (human figures and environments). Morphing tools? The Morphbrush is nice, but I prefer to build my morphs using a full fledged modeler and import it back into Poser - I've written a couple of nifty Python scripts to assist me there.
Rendering? Yes, I'd love to see some improvements in the Firefly engine. Rendering a decent water surface, reflecting metal or glass is sort of hit-and-miss magic with Firefly, while it's a breeze in Bryce, Vue or Carrara. I'd also love to be able to render more complex scenes - the full 64 bit renderer in Poser Pro is a step in the right direction, but I still can't render a crowd of medium to high-res people with clothing, hair, and a believable environment - which is easily doable in Vue.

The Hair Room. In fact, it's not bad at all. Easier manipulation of the hair for styling would be a big plus - named selections would be great. Better than hair groups, since named selections can be created on the fly, while hair groups must be planned in advance.

Cloth Room: I love it. In fact, I like it better than 3DS Max Cloth.
But it sure can use some improvements.

• easier selection of vertices (I'd love tools like "grow selection" and "shrink selection" comparable to 3DS Max subobject selection tools). Would be great for polygon selections too.
• multiple constrained groups, specifying WHAT object they're constrained to;
• multiple choreographied groups that can be keyframed independently;
• multiple soft decorated groups/rigid decorated groups, where one can specify what dynamic group they should follow;
• rigid decorated groups that are truly rigid (I haven't had much luck with those, buttons and the like tend to deform despite being designated as rigid decorated).
• pre-stretched cloth (think elastic bands);
• dynamic rope!!!! Would be a MAJOR plus.
• keyframed group switching: think buttons being undone and redone, think hanging a towel on a hook, think putting on/off a coat, knotting a belt, etcetera).
• last but not least: thickness of the cloth. Collision offset and collision depth specify the thickness for the simulation, but I'd love to see this same thickness in the render.

Setup room: please, please make it possible to let the first child bone have its origin at a different location than the parent bone's endpoint. And a precision of 3 digits in the numerical fields just is not enough. For precise manipulation you need at least 5 or 6 decimals.
Working with spherical falloff zones is also a regular pain in the neck. They always start out FAR too large and in a non-logical position. It would be better to spawn the zones right around the joint you're editing, with the inner mat sphere having a radius half the length of the current bone, and the outer mat spere having a radius equal to the length of the current bone. Those are far more sensible starting values.

Useful in more than just the setup room: a grid in the orthogonal views that can be snapped to. The grid units should be customizable to the units Poser recognizes (Poser Native, feet, inch, meter, centimeter, millimeter) with sensible subdivisions that appear when zooming in (and disappear when zooming out). A "snap to grid" toggle would be most helpful.

A "drop to underlying surface" button would also help. The current implementation of collision detection is far too slow, probably due to the insane amount of vertices/polygons that have to be included in the calculations.  (On a side note: "drop to underlying surface" can be implemented from Poser 5 upwards using Python. I've made a start, but I just haven't got the time to finish the script).

What else? Oh hes, the "more traditional user interface". If that means getting rid of the space consuming palettes and buttons in lieu of dockable toolbars/palettes, I'm all for it! And a hierarchical scene graph that allows grouping of multiple objects - much like the Hierarchy editor already does, but one that's more stable and responsive - would be greatly appreciated.

Oh, and don't forget the documentation. The current documentation is fairly hopeless. Yes, I understand that the "Bias" dial on a material node adjusts the bias. Putting that in the documentation is a waste of paper/screen space. What I want to know is what it DOES. What are sensible values? What effects does it have on the node? The primitive basics of Poser - those that haven't changed since version 3 or so - are laid out reasonably well, but the newer features are truly horribly documented.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff