why does cilk_spawn give me such bad performance?

I wrote a simple benchmark that fills two vectors with a sequence of randomly computed numbers. I'm comparing tbb, cilk, and std::async. The results I get are (milliseconds)

time cilk: 3725
time async: 2367
time tbb: 2397
time serial: 4610

I compiled with (icc 2016)

icc -std=c++14 -xHost -O3 -tbb -pthread benchmark.cpp

And I'm running on

uname -a
Linux localhost 3.10.25-gentoo #1 SMP Wed Jan 29 22:44:07 EST 2014 x86_64 Intel(R) Xeon(R) CPU E3-1245 v3 @ 3.40GHz GenuineIntel GNU/Linux

Why is cilk so much slower than tbb and std::async?