curr_connections shows incorrect value after huge amout of connection is established and disconnected

1. Start membase with max simultaneous connections as 65000.
/opt/membase/bin/memcached/memcached -d -p 11211 -u nobody -c 65000 -E /opt/membase/bin/ep_engine/ep.so -r -l 0.0.0.0 -e dbname=/db/membase/default;min_data_age=0;queue_age_cap=900;max_size=64424509440;initfile=/tmp/init.sql;ht_size=12582917;max_txn_size=250000

2. Make a note of curr_connections stats.

3. Start memslap from two different machines hitting the same membase server each with 31000+ connections.
./memslap -s 127.0.0.1:11211 -T 1 -c 256 --conn_sock=121 --time=1d

4. Run the above setup for 60 seconds and stop the memslap on both the machines.
Observed: The curr_connections stats comes down but does not reach the value observed in step 2. It will have 100+ connections.