Comparing async and sync benchmarks about benchmark.js

[bcherny]

The async test is calling a function that calls another function, so the return statement in it is at least 2 function calls deep. The sync test, on the other hand, doesn't call any functions.

A quick jsperf test case reveals that each nested function call roughly halves performance. So with basic naive function calls in IIFC's (to simulate a root function call in benchmark.js), we would expect a 4x reduction in performance, not the 300x reduction we're seeing.

Possible explanation: V8 optimizes those jsperf test cases because they are so simple, and for more complex functions (like deferred.resolve) the drop in performance is actually an order of magnitude greater than 50%. Looking through the bjs code there is no obvious difference between how a/sync code is run - it seems to be executed the same way after the compilation step. Hopefully @jdalton or @mathiasbynens can chime in to clarify.

[jdalton]

This is a dup of #39, see it for details.