Latency problem with 1k req/sec

I'm using 2.8.17 redis server.
Send requests throw unix.sock by nginx+lua. Each request is one SMEMBERS command. Now I have about 1k req per sec. In the nginx access log I see mostly next

[09/Oct/2014:18:04:49 +0300] [ 1412867089.670 : 0.002 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.670 : 0.002 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.670 : 0.002 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.670 : 0.002 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.671 : 0.001 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.672 : 0.000 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.673 : 0.001 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.673 : 0.001 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.674 : 0.000 : 198 ] 212.1.66.241 - - "POST     /API/0.1/url/getCategory/ HTTP/1.1" 200 37 
[09/Oct/2014:18:04:49 +0300] [ 1412867089.681 : 0.000 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 

You can see that request time is not more than 2ms. But almost every second I have a few requests with request time about 100ms

[09/Oct/2014:18:05:22 +0300] [ 1412867122.843 : 0.131 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:05:23 +0300] [ 1412867123.212 : 0.100 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29 
[09/Oct/2014:18:05:23 +0300] [ 1412867123.613 : 0.101 : 190 ] 212.1.66.241 - - "POST /API/0.1/url/getCategory/ HTTP/1.1" 200 29

I can't find the problem place.
I checked everything from the latency guide
Slowlog has no entries
System is not swapping
I don't use AOF

I read a lot of posts on stackoverflow like this
http://stackoverflow.com/questions/16841469/redis-performance-tuning/23719773#23719773
and I can't find the solution.

redis-benchmark gives me next

redis-benchmark -s /tmp/redis.sock -q -n 10000 -c 1 -d 50
PING_INLINE: 51282.05 requests per second
PING_BULK: 64935.07 requests per second
SET: 39840.64 requests per second
GET: 28901.73 requests per second
INCR: 26525.20 requests per second
LPUSH: 29411.76 requests per second
LPOP: 23255.81 requests per second
SADD: 21551.72 requests per second
SPOP: 23041.47 requests per second
LPUSH (needed to benchmark LRANGE): 45045.04 requests per second
LRANGE_100 (first 100 elements): 10101.01 requests per second
LRANGE_300 (first 300 elements): 4196.39 requests per second
LRANGE_500 (first 450 elements): 3337.78 requests per second
LRANGE_600 (first 600 elements): 7320.64 requests per second
MSET (10 keys): 62111.80 requests per second

Start with the nginx access-log samples and the Redis 2.8.17 setup using a Unix socket, nginx+Lua, and one SMEMBERS command per request. Reproduce the intermittent ~100 ms requests, compare them with the redis-benchmark results, latency guide checks, and empty slowlog, and identify and verify the source of the delay.