For Telemetry benchmark, I found that the first and last pages/stories' elapsed time include browser start time and measurement time, respectively.

For example: tab_switching.top_10 
[ RUN      ] https://www.google.com/#hl=en&q=barack+obama
[       OK ] https://www.google.com/#hl=en&q=barack+obama (45178 ms)
[ RUN      ] Wikipedia
[       OK ] Wikipedia (445 ms)
[ RUN      ] http://www.amazon.com
[       OK ] http://www.amazon.com (2231 ms)
[ RUN      ] http://www.yahoo.com/
[       OK ] http://www.yahoo.com/ (2492 ms)
[ RUN      ] http://www.bing.com/
[       OK ] http://www.bing.com/ (42100 ms)
[  PASSED  ] 5 tests.

You see that the first page and last page's elapsed time are order of magnitude longer than other pages. The first one includes browser starting time. And the last one includes tab-switching measurement time.

Shall we fix that?