Finally! After an eternity of revisions and attempts to reconcile the proportions of specimens of different ages that frequently didn't overlap, I feel confident I've nailed the general proportions.
With generous aid from several people ("credits" below) I figured out that the legs weren't actually too short in my original skeletal - in fact I ended up making them too long in my most recent attempt. Instead, the vertebrae and ribs based on UA 8678 (that would be the cervicals, dorsals, sacrals, and 5 anterior caudals) were scaled up too large.
That's an important distinction, as scaling them down had a chain-reaction on the relative size of the head, pelvic and pectoral girdles, etc.
It's still a strange and somewhat low-slung theropod, but nothing like my previous attempts. I should note that there weren't any mathematical scaling errors - rather, the way I chose to reconcile the axial series of UA 8678 to the larger tail and head of FMNH PR 2100 was wrong (the latter has proportionately taller neural spines, which is probably an ontogenetic feature).
Obviously newer data could require some nips or tucks, but I honestly think you are safe to illustrate Majungasaurus now. I know I plan to!
Special thanks (in alphabetical order) go out to Matt Carrano, David Krause, Adam Pritchard, and Scott Sampson, all of whom who provided excellent critiques and made this version possible (not to mention the research and publications the restoration is based on).
Now I'm going to go find a bottle of champagne to pop...