csdp duration grows with the number of cores available (12 is about 1000x slower than 1)

[Jason Gross]

Using the code from https://github.com/JasonGross/csdp-test (also attached), on my 12-core machine, I get a massive slowdown between using 1 core, and using 3 cores; the relevant snippet from the output of `make` in that repo is:

Testing on up to 12 cores
0 (real: 0.00, user: 0.00, sys: 0.00, mem: 3660 ko)
1 (real: 1.57, user: 1.56, sys: 0.00, mem: 3668 ko)
2 (real: 1.57, user: 3.10, sys: 0.01, mem: 3648 ko)
3 (real: 1.61, user: 3.17, sys: 0.01, mem: 3664 ko)
4 (real: 1.63, user: 4.78, sys: 0.00, mem: 3660 ko)
5 (real: 1.68, user: 6.44, sys: 0.00, mem: 3696 ko)
6 (real: 1.84, user: 8.60, sys: 0.00, mem: 3708 ko)
7 (real: 1.85, user: 8.82, sys: 0.01, mem: 3720 ko)
8 (real: 2.14, user: 12.07, sys: 0.02, mem: 3712 ko)
9 (real: 2.22, user: 12.59, sys: 0.01, mem: 3736 ko)
10 (real: 2.25, user: 14.65, sys: 0.01, mem: 3764 ko)
11 (real: 2.35, user: 16.65, sys: 0.13, mem: 3720 ko)

1)
$ lsb_release -rd
Description: Ubuntu 14.04.3 LTS
Release: 14.04
2)
$ apt-cache policy coinor-csdp
coinor-csdp:
  Installed: 6.1.1-1
  Candidate: 6.1.1-1
  Version table:
 *** 6.1.1-1 0
        500 http://us.archive.ubuntu.com/ubuntu/ trusty/universe amd64 Packages
        100 /var/lib/dpkg/status
3) What you expected to happen
12 cores should be no slower than 1 core
4) What happened instead
12 cores is about 1000x slower than 1 core

Note: Test cases from https://coq.inria.fr/bugs/show_bug.cgi?id=4983#c9