Running the attached test gives a feel of it (works fine with gcc on solaris):

[qt@kvasir:~/tests/task_fontconfig] time ./task_fontconfig 1 0
Multiline is 0 perftest is 1

real 0m0.173s
user 0m0.053s
sys 0m0.021s
[qt@kvasir:~/tests/task_fontconfig] time ./task_fontconfig 1 1
Multiline is 1 perftest is 1

real 0m2.469s
user 0m0.748s
sys 0m0.056s

This is apparently a regression from 4.4 (not tested with 4.4 though)