My own performance tests did not demonstrate any noticeable difference between Polylang 1.1.6 and 1.3 (I am aware that some performance was lost in some parts of the code, but other gains should compensate this). Note that these performance tests were done with almost no other plugins activated, so results may be different with different tests.
I would not use your way to get the performance impact of a plugin. According to my experience, P3 is very unreliable. If you want to measure speed using WP plugins, I recommend to use debug bar + debug bar extender (which includes a profiler).
To perform the comparison, you can choose a page and load it first with Polylang de-activated (look at the total execution time in debug->profiler and the total query time in debug->queries) and then remake the test with Polylang activated.
In fact, each test should be done several times as you will notice a big dispersion in results. Especially, you will want to remove the tests in with extreme results (if WP ran a wp-cron for example).
If you do your tests on a local Apache install, you can also use Apache Bench which can load the same page several times and will give you the avarage response time (here again, make the comparison with and without Polylang).
For your comparison to be accurate, you have to take care that the language loaded is the same with and without Polylang. English pages do not load any .mo file while other languages will load the corresponding .mo files for WP, your themes and your plugins. This has a significant impact on the page loaded time.
