Started rewriting md5loader. you can find it at http://gamedev.kproject.gr.
The code is in alpha stage, it works for single meshes and in order for animation to work
you must change every line in your md5mesh file that contains “mesh{” to “mesh yourmeshname {”.
Thank you!! 
two questions
- will it be able to blend animations and
- will it allow the use of bones for tagging/object hooks in the old loader in the same way you add the head to the body I utilize this and its offset methods to place weapons and accessaries on my models. e.g I have weapon, holster and accessary bones for my models used for attaching these. at present also rig my weapons with appropiately named bone/tag
I do like the naming concept in the mesh file though would allow for the
yes it will support animation blending.
for your second question since the meshes you want to attach are rigged then the only thing that is needed is to calculate the meshes position if it is not given.
Anyway i'm waiting for approval of jme-contrib dev.java.net project to continue development, and with a public cvs repository any help is welcomed and appreciated.
Personally i prefer the way cal3d handles animations. The long way plan is to define an abstract skeleton structure supporting animation blending and dynamic mesh attachment and provide the means to work with both md5 and cal3d (or any other skeleton based format) in a unified way.
the problem is I’m a hobbyist/novice and utilizing thebone hooks/tagging idea(picked from UT modding) allows me to escape some of the complicated math stuff.
e.g. I have models that utilize floating objects these objects are not put in the same md5 because these object will not always be part of the model as such I only animate the bone tag in some instances.
so is it reasonable to assume that there is no convenience methods (yet) I hope :’( for accurately placing the heads on doom3 flavoured models haven’t dug to deeply into it yet just read the test.
at any rate I’ll continue to use the old one until I figure out the new one or can nolonger fix the old one which ever comes first. 
://
can you provide me with a sample md5 model with attached meshes so that i can identify your problem and provide a solution?
I can toss together a simple example
Personally i prefer the way cal3d handles animations.
I don't know if it is better, but thing is...well, the exporters I have tested with 3d packages, are working great with md5 (blender is absolutely seamless, I don't use normal maps or anything more than smoothing, texture, uvmapping, skin and weights animation. It even graps perfectly the new pose mode, allowing to grab the interpolation, the ipo info. Needing not a key(bake) per frame, which also allows smaller size, over an already small way of storing. ), and mostly...I doubt an animation format can go lower in filesize per anim than md5...We're so going to base all character animation of our project in md5, as it allows me include a lot of animation data, without bloating the download size...
got to agree with snaga there, between cal3d and md5, md5 is the most strongly supported by blender at this time, cal3d was updated to work with newer blenders but still has bugs AFAIK. I dont use max or cstudio alot anymore except texture baking (mental ray kicks ass).
...besides, md5 is the only bones and weights format you can ouput from gmax, which is free (dunno how much time download will stay avaliable )
To date, md5 for me is the less problematic weights and bones animation export out of blender , even less problematic than other formats of this type, in other softwares!
Besides, is thought for be optimized for a game engine, since start (not doing a comparison with other formats, just speaking of md5) and certain coders have told me it's quite "prepared" for the real task. (basicly they told me Carmack knows his stuff ;) )
From a character animation point of view,(and any other ;) ) is eons better than md3.