All right. I have been thinking about this, programming-wise, for a bit, and here's the solution I came up with. Now, I do not know the full complexity of the current setup, but here's a way it could work:
I've seen a mockup in this thread for a new box on the right of the dress-up screen that lists the items equipped so that you can select their layer order. This can make the dress-up screen the source for the completed image, instead of having it re-rendered at other points in its use. Just have the dress-up screen capture the avi as designed, send that image to the cache servers, and have other processes go to those servers for the images. You could also have a flag of whether the avi has an animated image set, and when the flag is set, the dress-up app would render the additional frames of animation at the same time as the still, and other calling functions would read the flag and load the image as a single or multiple image as appropriate. "I Am" items already override regular items, so I see no reason why that would have to change. Just make the other items grayed out in the right hand z-level box for a visual indicator. Also, the "walk" animation in towns and zOMG is just a blurred leg swish, so that can be layered on top of anything. Finally, the back version of the avi can be set to the opposite of the front to begin with, but can be separately set as well. So, the final image set would be: front-right-facing, front-left-facing, back (flippable), f-r-kneeling, f-l-kneeling, back-kneeling, and when necessary, f-r-animation, f-l-animation, back-animation, f-r-kneel-animation, f-l-kneel-animation, back-kneel-animation, so either 6 or 12 images depending on animated item usage, and if you want a part of the site to not worry about the animation, just have it load the first 6 images.
As for layering within clothes, I'm sure there are some items that have more than one layer to accommodate backs to jackets that may be visible, and other such. What you could do is simply make the z-level of the item back layer a relative -1. You could further limit the regular item layers z-level, within the Dress-up interface, to a minimum of 1 or 2 to make sure that the relative -1 wouldn't break the code (and it doesn't need to be anything visible to the players, either).
So, in conclusion, make the dress-up program a little heavier, and you could actually lighten a bunch of the other processes, probably save on some server resource usage, and it would standardize the avi's look across the site as well!