In the spirit of working out inflection points and so on. Here's some results for a series of simulations done with the java simulator Nils mentioned early on. This will take a while to explain.
http://melee.psyops.org/round1.png
This figure plots the results of pair-wise duels for the full range of 32-point figures (ST8 DX16 all the way to ST16 DX8), and for varying levels of armor. This is strict Melee, no talents or IQ shenanigans. This is from years ago, and it is NOT using the new modified plate armor penalties. Shields are in two sizes, small and large.
On the x-axis is ST for the figures. Figures are always using a weapon requiring their full ST. On the y-axis is their adjusted DX. Because the sum of ST+DX=24, you can pretty well infer from the adjDX the amount of armor being worn. Note that figures with 1H weapons are always using a shield of some kind.
The color indicates the win-rate for that figure against all opponents.
So as an example, you can see that ST12 figures wearing light or no armor are doing well overall. In fact wearing lots of armor hurts them.
On the other hand, ST8 figures without armor are terrible, and improve somewhat by wearing lots of armor.
As a guide, the blue line indicates the minimum armor setup (just a small shield if using a 1H weapon, or none at all if using a 2H weapon at ST14+). The maroon line is the maximum armor setup (plate and a large shield if 1H, or just plate if 2H).
There are some idiotic results here. The ST16 fighter in full plate (adjDX=2!) is winning quite a few battles. These are of course against very light fighters. Whereas he just needs one automatic hit to win, they need multiple 2x and 3x hits to penetrate his armor.
If you limit the competition to the best of these fighters, and have them fight just each other, then these kinds of peculiars fall away, and you start to converge on the best overall fighters. The final round results are here:
http://melee.psyops.org/round3.png
Basically, ST11-13 and light armor usually produced the best results. This won't be very surprising for experienced Melee'ers.
I offer this just as one way to think about digesting the large amounts of data a simulation can generate.