inaccurate vb_active_ht_memory value reported in a cluster after few rebalance operations ( replica=2)

vb_active_ht_memory and vb_replica_ht_memory dont seem to add up
in an environment where
number of items = 190 million and key size is around 50 bytes the vb_replica_ht_memory should be near 48 GB but the value reported is much less ( 1.3 GB)
aggregated stats: ( from Moxi)
vb_active_ht_memory: 1381449232
vb_pending_ht_memory: 0
vb_replica_ht_memory: 2791803296
[root@mem4 ~]#

per node stat :
[root@mem4 ~]# /opt/membase/bin/mbstats localhost:11210 all | grep ht_
vb_active_ht_memory: 163441408
vb_pending_ht_memory: 0
vb_replica_ht_memory: 362664672

there are 8 nodes in this cluster so the vb_active_ht_memory should be near 6 GB in each node but memcached reports 163 MB