They completely rewrote the support generation part of it and sped it up immensely.  It is a huge improvement over the 0.9.X versions with broken support generation.

As far as manual support generation through CAD goes, that's an interesting idea as I use SW at work and am very proficient with it. I have used it already multiple times to rework STL files so that they are easier to print.

You can open STL files in SW and change the import settings such that SW generates "dumb" geometry when importing it. You can then use them noramlly and modify to your heart's content.

OK...so Alex on the Github posts say it's fixed...how does a coding luddite like myself apply the fix?  I could not even locate the "fixed" file...

Like myself, you wait until Alex releases 0.9.11, which by the looks of it is way off, or beg for a compiled version by someone around here that knows how to do that .  Both 2n2r5 and Adrian hinted that they might be able to do that in a different post I have on this subject...

http://www.soliforum.com/topic/4149/slic3r-v0911/

..but for now we wait.

It was suggested that I downgrade to a previous version, and I did, down to 0.9.9, but it doesn't work there either.

When Alex closed the open issue on Github, he suggested the fix was around the corner.  That was a month ago.  At that time the Milestone showed 75% completed.  AS of now that number stands at 76%.  At this rate, we're looking at about 2 years for the release...

It's also very wasteful, depending on the amount of support it generates, you could have more plastic going into the support than in the model itself. 

Here is the example I posted on GitHub, model itslef is only requires about 3200mm of filament, while the supported model is in excess of 7200mm more than twice.

Edit: And, by glancing at source, it seems my guess was way off... For one, threshold applies to angle between layer edges not triangles -- and generally everything is done on toolpaths, not the original mesh?

Edit2: Not near computer, may give STL shot over weekend..

Ah:
https://github.com/alexrj/Slic3r/issues/1082

Broken in 0.9.10b, fixed in master..

Well, that's nice, but the "master" hasn't been posted in binary format, and since I'm not a programming guy I woudn't know how to compile the new version from the source code.

Do you know of any source for the latest binaries?

Edit:

Here is the STL model of the Little Pony

Broken in 0.9.10b, fixed in master..

Last part I tried gave me some trouble.. changing angle even by 2 deg and spacing by .05 made a difference between no support or half-missing support, and decent support.. and that was on a completely flat 180 deg surface (a bit thin and long, though). So.. not intuitive, definitely.

Try angles greater than 90. No support is 180 (think of the angle of a tangent to a circle with axis parallel to x-y plane: top is 0 deg, sides are 90 deg, bottom is 180 deg). That said, I still cannot get it to do exactly what I want. My guess is that slic3r first prunes triangles whose tangent is below threshold, then projects remaining triangles onto x-y plane to find support area, then uses spacing in some way to decide where and how to place support over these areas... So the interplay between these parameters (angle and spacing) is non-intuitive. I'll have to "use the source" (github repo) to see if my guess is correct...

I already tried that.  Anything over 90° generates very little support, usually in the wrong place or only at extremities, adn sometimes the support is incomplete (like it never reaches the part it's supposed to support).  Not only that, but sometimes it takes like 3 times longer to generate the code.  Not very useful in my opinion.

One thing I found useful though is the option to generate support for the first ## layers even when the support checkbox is disabled.  That seems to generate support for the first 50 layers or so, and it works pretty well (like when you have a bottom surface which is not flat or it's slightly above the build surface.  That works OK.  You can input any number you want, but the support seems to top out at around 50 layers.  Additionally, it will only place support that starts on the build platform, and no support above the actual model surface

Try angles greater than 90. No support is 180 (think of the angle of a tangent to a circle with axis parallel to x-y plane: top is 0 deg, sides are 90 deg, bottom is 180 deg). That said, I still cannot get it to do exactly what I want. My guess is that slic3r first prunes triangles whose tangent is below threshold, then projects remaining triangles onto x-y plane to find support area, then uses spacing in some way to decide where and how to place support over these areas... So the interplay between these parameters (angle and spacing) is non-intuitive. I'll have to "use the source" (github repo) to see if my guess is correct...

It would be fun to compare the algorithm for support in skeinforge and slic3r. It always seemed to work a bit better in skenforge.

... even if it's just to berate them for not doing their due diligence in searching.

I find the beratings sadistically amusing, keep up the great work everyone!
If folks are too lazy to search for themselves, they should deserve/except to get called out about it in a witty [usually] tactful manner.

What that saying.... "There's no stupid questions, only stupid answers." :-)