How to use"fix gcmc" for the relaxation of H-atoms in metals?

The key is to see what is happening to the H atoms. I suggest you do a small test simulation, running fix gcmc every timestep, and dumping atom positions, velocities, forces every timestep. My best guess is you are losing atoms because the inserted H atoms have large forces, putting the timestep beyond the numerical stability limit. This could be happening very infrequently or every time.