Eh, I feel a lot of the problems with us hitting the BMP limit is that we're just not very smart when it comes to deciding what textures need to be resident versus what doesn't need to be resident. A lot of that is outside the internals of bmpman and more of determining what needs bm_release to be called on it.
The other problem is that the way we do animations, especially for effects, is kind of ridiculous. Modern game engines atlas most of their effect animations. Here, we're still using series of individual textures. Most of Wings of Dawn's textures come from small effect animations and had Spoon had the capability to shove all his effects into atlases, we wouldn't be having this conversation right now (Maybe later once a bigger structural issue with BMPMAN is revealed)
Regardless, I feel that if we are to scrap BMPMAN for something else, we need to keep in mind of how modern GPUs handle textures. Even now we're using depth render textures, stencil targets, floating point textures, texture buffer objects, and array textures. The scene textures and g-buffers that are used for post processing, deferred rendering, and soft particle rendering are completely divorced from BMPMAN. I'm trying to think of ways to pull them back in but it's going to require some thinking on my part. If any actionable discussion comes from this, I'm going to suggest you guys wait until I get a good idea of what needs to be done.