These benchmarks were performed on a virtualized Rackspace Cloud server running Ubuntu 13.10, with the "30GB Performance" flavor (8 vCPUs, 30GB RAM, 2.5 Gb/s network)
GHC (compiled from git sha ec4af3f):
$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 7.7.20140107
Server
$ ./dist/build/server-threaded/server-threaded 5055
Listening on 5055
Client benchmarks
$ time ./dist/build/client-single/client-single 127.0.0.1 5055 1000 100000
Connecting to 127.0.0.1 5055
conc: 1000 reqs: 100000 batchSize: 25
100000 successes, 0 errors in 9.024822 s
min/max/avg request time: 91.07590 us / 7.280918 s / 20.69674 ms
11081 r/s
real 0m9.058s
user 0m2.183s
sys 0m6.874s
$ time ./dist/build/client-threaded/client-threaded 127.0.0.1 5055 1000 100000
Connecting to 127.0.0.1 5055
conc: 1000 reqs: 100000 batchSize: 25
100000 successes, 0 errors in 8.152756 s
min/max/avg request time: 166.8930 us / 7.014485 s / 47.34623 ms
12266 r/s
real 0m8.200s
user 0m11.045s
sys 0m12.633s
For reference, here's how the same code worked with GHC 7.6.3
$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 7.6.3
Server
$ ./dist/build/server-single/server-single 5055
Listening on 5055
Client
$ time ./dist/build/client-single/client-single 127.0.0.1 5055 1000 100000
Connecting to 127.0.0.1 5055
conc: 1000 reqs: 100000 batchSize: 25
100000 successes, 0 errors in 8.877150 s
min/max/avg request time: 58.88939 us / 1.060330 s / 1.040994 ms
11265 r/s
real 0m8.940s
user 0m2.200s
sys 0m6.735s
$ time ./dist/build/client-threaded/client-threaded 127.0.0.1 5055 1000 100000
Connecting to 127.0.0.1 5055
conc: 1000 reqs: 100000 batchSize: 25
100000 successes, 0 errors in 14.46206 s
min/max/avg request time: 190.9733 us / 7.046602 s / 137.7901 ms
6915 r/s
real 0m14.545s
user 0m17.072s
sys 0m20.840s