Alot of your math is overly complex. You should seek to create simpler versions that have the same feel.
Ie, your armor table:
Max dex = 8 + strength mod - AC/2
ACP = 0 + strenght mod - AC/3
Either do away with armor penalty, or generate a table based off of total AC.
Now, all you need is AC by type in each slot (to line up visual effects with game stats), and the other values "fall out" for free.
Padded/Silk: 1 AC per slot (note: silk is super expensive)
Leather: 2 AC per slot
Metal studs on Leather: 3 AC per slot
Hide: 3 AC per slot
Chain: 4 AC per slot
Piece Plate: 5 AC per slot
Full Plate: 6 AC per slot
Each of the above can be made "light" or "thin" for -1 AC in the slot.
Your save costs seem off.
Both your save and skill costs should be streamlined. There is little need to have the cost of something be determined down to 3 decimal digits -- that is a sign you need to clean up the mechanic.
One idea is the following:
base + mod
Mod goes up every (ability modifier that determines talent) ranks in the skill/save. Negative in your talent-skill adds a one-time boost to the mod cost, and make it go up 2 steps every notch.
Assuming base 100 and mod of 10, a price/talent matrix:
Or some tweak on it -- the goal is to reduce the number of significant digits involved.
-2 -1 0 1 2 3
1 160 140 120 110 110 110
2 180 160 140 120 110 110
3 200 180 160 130 120 110
4 220 200 180 140 120 120
4 240 220 200 150 130 120
4 260 240 220 160 130 120
4 280 260 240 170 140 130
Next, why not standardize your feat costs? Just make the sub-par feats a bit better, or take the strong feats and split them up.
For combat, your system doesn't encourage a very simple stereotype -- the "guy who is good with weapons". Instead you have people who are good at particular weapons, and being good with many weapons is next to impossible.
Knowing a second weapon type after you know a first isn't a huge advantage -- it is more of a flavour thing than anything else (especially given how bulky weapons are).
Lastly, your system ignores the fact that the method and style of fighting matters more than the weapon in many cases. A greatclub and a greatsword fight more similarly than a longsword and a greatsword does.
To steal from a different game, how about a few basic skills:
+ Melee (con and wis)
++ Style (2H, S+B, TWF, Duel, Close)
+++ Weapon Types
+ Ranged (dex and wis)
++ Style (Close, Medium, Volley)
+++ Weapon Types
Both your Style and your Weapon Type skill are cheaper than your Melee or Ranged skill. They are capped at 1/2 of the Melee or Ranged skill.
Your ability to use a weapon is the sum of all 3 parameters.
Now building someone who is good at fighting with their fists is a (high melee) (high close) (high twf) (high fists) (high monk) skill character.
Meanwhile, a knight might know (high melee) (high S+B) (high 2H) (high sword) (high lance) (high shield).
If the Knight picked up a mace with a shield, she would be at 3/4 of max skill. The Knight fighting without a shield, using a 1 handed sword, would also be at 3/4 of max skill.
Note that the weapon types need not be exclusive, if you know "monk weapons" and "staff weapons", you only get to use the highest when using a quarterstaff.
Next, might I strongly encourage using the same base system for saves/skills and weapons. At most have a cost multiplier next to them.
This means that players only have to learn one system for all of the rank-based advancement mechanisms, and means that "bugs" will get ironed out faster rather than fester in a dark corner of the system.
Your system is probably fine, I just thought I'd toss out comments on the rougher spots. :)
IMHO, a fantasy gaming system should aim for:
1> Decision resolution.
2> Ability to create traditional archtype characters.
3> Math that is simple to do.
4> Fewer special-case rules, the better.
Rules that are broad and general are good. When in doubt, try to fit a rule into an existing mechanism, and see if it works.