KK2 diagonal wobble on x quad

[Jeez]

I've got an issue that I can't work out, hoping for some new insight.

My quad is a symmetrical (no dead cat like setup) x configuration. The arms left front and right back react really snappy, but the other two overshoot like crazy. This is most noticeable on moving diagonally and yawing quickly.
The motors take longer to spin up, and power down after that. When shaking the quad while hovering it snaps right back to level when moving the left front and right back arms, but the other two make it jump to the side like crazy.

So far I've tried and done the following:

• Eliminated vibrations
• PI tuning
• Calibrated accelerometers
• Done a sensor test
• Reflashed ESC's with simonK (had done that already, but just to be sure)
• Changed/rotated ESC's
• Changed all wiring
• Changed motors from Turnigy to better quality motors and carbon props.
• Updated KK2 software from 1.6 to 1.6++ rev3 (and went back to 1.6 because flying in acro caused the level position to drift on the newer firmware, so much for that failsafe)

Unfortunately, I don't have a spare KK2 to swap. The only explanation I can come up with for a cause is the gyro/acc not functioning properly.

Am I missing something? I like flying aggressive FPV, and this is making it highly unpredictable.
Should I save up for a new KK or would a CC3D be a better option for acro flying? I need my quad to be as "locked in" as possible.

 

[Mustang7302]

How is the board mounted?

 

[Jeez]

I knew I forgot some detail somewhere
It's mounted on a double layer of gyro foam pads.

 

[Mustang7302]

Pictures of how things are mounted? List your P & I gains/limits for the three channels. Is pitch and roll gains linked?

 

[Jeez]

It'll be a couple days until I can make pictures. It's an RCTimer frame kit: http://www.rctimer.com/product_1055.html.
Everything is screwed with loctite, esc's are zip tied to the frame. Battery is on top, KK2 is on the bottom plate at the CG.

Roll/pitch linked. Yaw settings are same as this. I have tried all kinds of values, but until I can make all motors react the same it won't matter. It remains unstable because of the differences. Higher gains cause the quad to fluctuate in altitude more violently.

PGain 50
Plimit 100
IGain 25
ILimit 20

 

[Cyberdactyl]

When you go to the 'Sensor Test', do you get 'ok' for all six readouts?

 

[Jeez]

Yes