I really like the direction you went with the last picture here, the value study or highlight, halftone, and shadow. Breaking things down into those base pieces/planes, especially when doing initial thumbnail sketches of an image, can really help, or at least does with me.
If you'll bear with me, I'll make a few suggestions of things that are helping me out. You might try creating a few thumbnails of the above objects, in their simplest forms (so, the outline minus details). Squint until the object (including portrait) is blurry so you can only see what's dark shadow and what's not. Using the info gathered from squinting, map the dark shadows onto the sketch. Then fill those shadows in. Do that for each. Now, create another sketch of the same objects, but this time, squint just enough to find the darkest shadows. Map the darkest shadows, fill those in. Go back to squinting and find the halftones. Map the halftones, fill those in. Leave the highlight and lightest points white. Compare those small, simple thumbnails to the original images in this last post.
Now, if you aren't insanely bored, consider doing another, larger sketch, where you map out the highlights, halftones, and shadows, then ease in the details as you fill in those values. All the objects we are creating are supposed to mimic light reflecting off a form, or being prevented from reflecting off a form (like cast shadows). Breaking the objects into planes, simple shapes, and forms, and mapping out the shadows can really save time on rendering and leave you more time to add in the details. Of course, as GaussianRaider recently exhorted me, shadows should be flat, and details put in the light.
Also, as ja said, do a lot of imaginative sketches. When you do those without reference, it'll very quickly reveal gaps in your knowledge. "Dear God that arm looks terrible." "Why does their waist look so waspish?" "Um... what does a lion's head look like again?"
I hope this helps. You are improving and I'm already subscribed to the thread.