That's what I've been doing, but it breaks contigious meshes because each "island" you create ends up seperate from the other ones.

If you're really keen you could sign up to the developer mailing list and start a discussion about improving things, but unless you're also prepared to write the code I wouldn't expect much action.

Currently the operator only cares about vertex location. It it should be taking into account weightmaps and normals too, and should also look at face normals to ensure that it doesn't remove back-to-back polys.

It also ought to be more consistent when preserving edge flags (e.g. sharp). That was my original request...since then BMesh has been merged in, so it might be easier to support now.

I've asked for improvements to the operator before but there wasn't much of a response. Feel free to make your own attempt. :)

Unfortunately I've not been able to fix this as I don't have fine enough control over Blender's "Remove Doubles" operator. Anything I did that worked with your example file (two seperate objects) broke importing contigious objects with overlapping parts and vice versa.

What I've done instead is add a new import flag which skips the Remove Doubles step entirely, allowing the user to fix the problem manually. Here's an unofficial release.[steamreview.org]

I can't reproduce the problem. Which version of the tools are you using?

Please also upload your blend file: http://www.pasteall.org/blend/

That sounds very wrong. I'll look into it.

Plans to try and convert normals to sharp edges have been shelved given that work is currently underway to add full support for custom normals to Blender. Once that arrives it will (unsurprisingly) be trivial to import normals.

The new smoothing system is still based on marking edges as sharp, so didn't change anything with regard to importing.

Face normals are compared too.

I don't need any files to fix it, and unless something more urgent crops up it will be in the next release.

There is actually already code in the SMD importer for this situation. It's having an effect (try selecting all verts then running remove doubles yourself!) but evidently more work is needed. Comparing weightmap data as well as vertex location should sort the problem out.

It wouldn't be anywhere else. :)