There are a couple of things I managed to figure out;
Watching Edy's video I realized my car is making too much wheelspin, 570Nm and a gear ratio that spins the wheel at 120Km/h in first gear. (Thanks Edy :) )
I tried with "less powerful" setup and voila got much better.
My keyboard control was also too slow for drifting and took too much time to countersteer, that also didn't help get things better.
Yet, my pacejka combiner is still weird, I doublechecked the original Beckman's paper and everything seems ok, the situation is the following;
Starting the "drift maneuver" the rear end gets some angular momentum, and when it start skidding (still spinning tires) because the slip ratio is high the lateral force "disappears" and I'm unable to stop spinning out, and while the engine is revving at max it takes about a second to get back the traction.
Edy, would you like to discuss your pacejka combining method? ;)
Here is what I use:
Normalized slip parameters
s=slip_ratio/optimal_sr
a=slip_angle/optimal_sa
Normalized slip functions
Fx(optimal_sr) = Ox(1)
Fy(optimal_sa) = Oy(1)
p=sqrt(s^2 + a^2)
Combined forces:
Fx(s,a)=(s/p)*Ox(p)
Fy(s,a)=(a/p)*Oy(p)
With taking care of original and cimbined slip signs, it should be the famous Beckman's model :)