Logstash pipeline bench report

logstash-pipe-bench.txt

For all below, this applies...
ruby -v
jruby 1.7.20 (1.9.3p551) 2015-05-04 3086e6a on Java HotSpot(TM) 64-Bit Server VM 1.8.0_45-b14 +jit [darwin-x86_64]

1) As a baseline for experimentation I substitute the JSON parser and generator for 'perfect' simulations.
   So for parse I simply return a new Hash and for generate I return a new String with 1 interpolation.

yes '{"test": 1}' |  bin/logstash -e 'input{stdin{codec => json_lines}} filter{clone{}} output{stdout{codec => json_lines}}' | pv -Wlart > /dev/null
01:00 [89.3k/s] [  89k/s]

  This give me a 'high water' mark.  Even with a perfect JSON parser/generator we can't go faster than this.
  
2) With JrJackson at v0.2.9 (current).
yes '{"test": 1}' |  bin/logstash -e 'input{stdin{codec => json_lines}} filter{clone{}} output{stdout{codec => json_lines}}' | pv -Wlart > /dev/null
01:00 [54.2k/s] [50.9k/s]

3) With JrJackson at v0.3.0 (new unreleased)
yes '{"test": 1}' |  bin/logstash -e 'input{stdin{codec => json_lines}} filter{clone{}} output{stdout{codec => json_lines}}' | pv -Wlart > /dev/null
01:00 [73.1k/s] [71.4k/s]

BUT Here is the strangeness!

If I remove the JSOM lines codec from either end, I see a reduced thoughput.

For perfect JSON...
yes '{"test": 1}' |  bin/logstash -e 'input{stdin{codec => json_lines}} filter{clone{}} output{stdout{}}' | pv -Wlart > /dev/null
01:00 [56.1k/s] [54.6k/s]

and...
yes '{"test": 1}' |  bin/logstash -e 'input{stdin{}} filter{clone{}} output{stdout{codec => json_lines}}' | pv -Wlart > /dev/null
01:00 [84.9k/s] [  85k/s]

I see the same relative numbers for old JrJackson and new JrJackson when removing the JSON lines codec from either end.

Question:
What is the explanation for this?