REST2 metadynamics -> convergence interpretation help

[peter.rich...@gmail.com]

I run a REST2 simulation combined with a well tempered metadynamics using the CV alphaRMSD.

Judged by the free-energy difference of two basins using the "analyze_FES.sh" script the simulation is away from convergence after 290 ns.

I resorted the trajectory by "trjcat ...... -demux replica_index.xvg" as suggested by the "Replica exchange I" tutorial:
http://plumed.github.io/doc-v2.1/user-doc/html/belfast-7.html

so that each starting structure regardless of its position on the replica latter is converted in one trajectory. The "rerun" of three replicas results in the figure presented in the attachment. (procedure as in the tutorial)

The main question is if the convergence is a question of time or if the set up is somehow wrong in my case?
Is the clustering of the CV values in the figure a sign that the well tempered ensemble is decreasing the HEIGHT of the metadynamic Gaussians to fast (wrong BIASFACTOR?)? 
Should the figure show a homogenous distribution of the CV values if the simulation comes close to convergence?

I appreciate any comments!

All the best
Peter




Further information of the simulation:
The 9 replicas have the effective temperatures:
300 320 341 363 387 413 440 469 500

The plumed input is:
######### plumed.dat.0  ##########################################################
MOLINFO STRUCTURE=struct_Renamed.pdb
ALPHARMSD RESIDUES=all TYPE=DRMSD LESS_THAN={RATIONAL R_0=0.08 NN=8 MM=12} LABEL=a1

METAD ...
LABEL=metad
ARG=a1.lessthan
PACE=500
HEIGHT=0.7
SIGMA=0.1
FILE=HILLS
GRID_MIN=-0
GRID_MAX=8
GRID_SPACING=0.1
BIASFACTOR=15.0
TEMP=300.0
... METAD

PRINT STRIDE=500 ARG=a1.lessthan FILE=COLVAR0
################################################################################
Each replica has its own plumed input and vary by an increased HEIGHT=0.7 to HEIGHT=1.17


The mentioned figure of the CV values of the resorted replicas as well as the free energy profiles are attached.