In my mind it makes them too strong, likely should be balanced with other stuff.

I think 20% + level 20 gem its something like 50% penetration. Which is huge means. if you had 75% cap'd resists before you would be taking 3x the damage. No way for the defender to stop this.

Example
100 base Cold damage vs 75% resist = 25 damage taken.
100 base cold damage +50% Cold Penetration vs 75% resist = 75 damage taken.

So even if the defender had 150 Cold resist with 75% capped, he would still take 3x the damage.

Unlike Ele weakness, you can keep capped resist by stacking a resist.

agreed, it blows every other support gem out of the water for any elemental skill

Except everyone already uses resistance curses, and the penetration gems are decently expensive. They're powerful, but at the end of the day they're far from mandatory.

Negative maximum resists is tons more valuable than any of the Resist curses. Anyone who's not using -Max resists in PvP is just shooting themself in the foot with a tank shell.

Example
100 base Cold damage vs 75% resist = 25 damage taken.
100 base cold damage +50% Cold Penetration vs 75% resist = 75 damage taken.

Invariably people will overstack resists to counter Resist curses; that will keep them at 75% Resists (or more). Even going from 75% to 50% means taking twice the damage, and the Penetration supports will make that basically unavoidable. There's nothing that comes close to more than doubling your damage output that reliably.

For PvE, it only matters against monsters with very large amounts of overstacked resistances; Res auras and the likes. It also doesn't require casting, making your clears faster. Won't be mandatory for PvE, but they're good nontheless.

"Adds X to Y Damage" looks bad.
Ascendancies were an interesting stepping-stone towards Atlas of Worlds. Make characters more powerful -> give that power a purpose. Might actually be less annoyed by the blatant power creep if Atlas works out.

So I've been asked for details of evasion over PM, and don't want to have to keep giving single replies to more people in future, so since I don't think it's all been laid out in one place yet, it will be here.

Evasion in PoE is not fully random.

Each entity in the world contains an 'evasion entropy' value, between 1 and 100. The higher this value is, the more likely they are to be hit by the next attack. The initial value is random.
Every time something attacks you, they calculate their chance to hit as a percentage. That value is added to your evasion entropy. If the result exceeds 100, you're hit, and 100 is subtracted from the value. If the value hasn't reached 100, you're not hit.

Before anyone starts clamouring that they're not getting their actual chance to hit/evade, let's examine this mechanic in a bit more detail. Take the simple example of 100% chance to hit. Since you always add 100 to the entropy, it'll always exceed 100, and thus always hit, which is correct. The case of 0% chance to hit can similarly be trivially shown to be correct.
So let's look at 50% chance to hit. Since the initial value is random from 1-100, there's a 50% chance that the initial entropy value is higher than 50%, in which case adding the 50 from chance to hit will exceed 100 and thus hit, and a 50% chance the value is 50% or less, in which case adding 50 will not exceed 100, and thus not hit. So the first hit has a 50% chance to hit, as it should.
The second hit also has a 50% chance to hit, but will never hit if the first one does - provided you're only getting hit by things with 50% chance to hit you, you'll evade every second attack, and be hit by the others.
Let's say the initial entropy was 42. The first hit increases this to 92, and misses. The second raises it to 142, hitting, and then subtracts 100 from the value, leaving it back at 42.
I'll leave other percentages as an exercise for the reader, but they all work out - if an attack has 25% chance to hit you, every fourth attack will hit, and so on.

This is the mechanic by which streakiness is removed from evasion - it removes the possibility of failing to evade happening to come up several times in a row due to bad luck. Each attack has the correct chance to hit, and will hit you just as often as you'd expect in the average case using a purely random system, but the possibility of occasional but devastating non-average results - such as being hit by four consecutive attacks with only 10% chance to hit each - have been eliminated.

Some caveats:
1) If an attack would crit you, evasion is tested a second time, and if you evade, the hit is downgraded to a non-crit (it does not miss, since it's already tested for that and hit). This roll is purely random and does not increase the entropy value - it just generates a number from 1 to 100 and compares to the chance to hit. Details of why are in the spoiler.

Spoiler

Given that this, if it occurs, would always be the next evasion test after the one to see if it hit, then if this did use the entropy value, then having above 50% chance to evade would make you immune to critical strikes, since you can't fail two successive evasion checks on entropy if their chance to hit is below 50%. If you were hit, that means you just failed the evasion check to evade the attack, and thus the entropy is such that the second test would be unable to hit you, and the crit would downgrade, whereas if it failed to hit in the first place, then critting or not is irrelevant.
While the concept of being so evasive you can't be crit is cool, the above behaviour is undesirable, and so checking chance to hit for the purposes of confirming a crit should actually stay a crit does not test entropy. Testing chance to hit for the purposes of actually hitting is always done via the entropy value.

2) Whenever the entropy value would be used, if a certain short amount of time has passed since the last time this occurred, a new random initial value is chosen. This prevents the player from waiting near a weak enemy until it hits (leaving them on a low entropy value), then running to a boss fight, to start knowing they'll have the maximum number of attacks evaded before they get hit. Entropy will perform it's function as long as you're continuously being attacked, but don't expect to transfer it from fight to fight.

well its random beforehand but fixed during battle and only during. i can live with that.
so by that pattern, if an attempt to evasion fails (token not add up over the specified eva value), then block and dodge have thier own constantly cycling entropy but only happening if the previous damage dodge failed.

so by that pattern, if an attempt to evasion fails (token not add up over the specified eva value), then block and dodge have thier own constantly cycling entropy but only happening if the previous damage dodge failed.

No. Only evasion uses the entropy value. Dodge and block are completely (psudo)random.