Postgres stream replication performance slow

目前有两台postgres db, 使用stream replication 做主从同步，在比较重的写操作时发生了比较重的延迟．

通过对　
pg_xlog_location_diff(pg_current_xlog_location(), sent_location) AS not_sent,
pg_xlog_location_diff(sent_location, write_location) AS not_written,
pg_xlog_location_diff(write_location, flush_location) AS not_flushed,
pg_xlog_location_diff(flush_location, replay_location) AS not_replayed
FROM pg_stat_replication;
这几个参数的监控，我们发现在
pg_current_xlog_location() - (sent_location)有比较重的延迟，延迟的size将近在100MB 甚至更大．
网络的吞吐量最大值只达到16MB/s.

本地模拟测试，增加数据表的字段　
一个表将近1300W数据，size 在10G.
我们在观察主从两边的wal　log 的数目时发现，从库远远落后于主库，整个操作会生成104 个wal 文件．
手工对比主从的wal file. 从库将近落后３０～５０多个wal file.

我也使用了iperf去测试了本地网络情况

Client connecting to 192.168.6.55, TCP port 5001

TCP window size: 125 KByte (default)

[ 3] local 192.168.6.56 port 54328 connected with 192.168.6.55 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 1.09 GBytes 939 Mbits/sec

1GB 已经达到了网卡的极限．

我们数据库所用的包为官网的包．Max_send_size = xblocksize * 16 (xblocksize = 8k default value)

问题是：理论上stream replication 可以达到我网卡的极限值，即１Ｇ/s，所以这次同步的延迟应该保持到１s以内．
有没有其他参数会影响postgres的stream replication.
期待高手的答复．