[Wien] convergence of slab calculations

[Kakhaber Jandieri]

Dear Wien2K users,

I am trying to relax GaAs surface with pair of nitrogen impurities.
For that I construct the supercell with 96 atoms ( Spacegroup: 6_Pm. No
inversion symmetry).
In total I have 12 layers with 8 atoms in each.
First (bottom) layer consists of hydrogen atoms (for passivation).
Next two layers are fixed.
In the last (top) layer two As atoms are replaced by nitrogen atoms.
I generated k-points with following  divisions of the reciprocal lattice
vectors:
10,1,1 (My slab is in x-direction).
My case.klist file looks as follows:
         1        10        10         1        20  2.0 -7.0 
1.5         0 k, div: (  1  1 10)
         2        10        10         3        20  2.0
         3        10        10         5        20  2.0
         4        10        10         7        20  2.0
         5        10        10         9        20  2.0
I do the calculations with MPI+k-point parallelization using 20 processors.
Here is my .machines file:
granularity:1
1:node132:3 node129:1
1:node129:2 node124:2
1:node124:1 node136:3
1:node120:3 node126:1
1:node126:2 node134:2
lapw0: node132:3 node129:3 node124:3 node136:3 node120:3 node126:3 node134:2

My conversion criteria are:
min -NI -i 10 -s 2  -j 'run_lapw -p -I -i 40 -ec 0.001 -fc 1.0 -cc 0.001'.

Energy tp separate core and valence states is  -6.0 Ry.
RKmax = 7.0.
I use the following values of RMT:
RMT(Ga,As)=2.0
RMT(N)=1.68
RMT(H)=1.5

The calculations already took almost the month.
Six mini cycles are already passed.
The energy decreases.
The forces also decrease for all atoms except of those in fixed and
passivating layers.
In these layers the forces are extremely large  (around  80)  and 
continue to increase from  cycle to cycle.
I already don't believe that the supercell will be relaxed at all
(because of these fixed and passivation layers).
Could someone give me the advice.
Is such behavior normal ? if not, what is wrong with my calculations?

Beforehand thankful,

Kakhaber Jandierei.