Thread creation performance

I noticed that thread creation performance has gone
down quite a bit compare to earlier kernel like 2.5.42.
I am running "test_str02 -l" (an NGPT test program) to
create maximum parallel threads using NPTL on RedHat 9.
It has glibc- and nptl-devel at 2.3.2-27.9 and 
kernel-2.4.20-8. I also tried with 2.5.66 kernel and
that's even worst. 
This test creates about 32700+ threads in 3m:27s, it
used to do it under 4secs. See the similar discussion:  
https://listman.redhat.com/pipermail/phil-list/2002-October/000087.html

I checked that both 2.4 and 2.5 kernel has Ingo's patch in it:
https://listman.redhat.com/pipermail/phil-list/2002-October/000083.html

Anything changed in kernel recently? or what else could be the problem?

Thanks, 
Saurabh Desai