Ok,

 first try, http://dforce.sh.cvut.cz/~seli/download/malloc-kde.tar.bz2 .
 There's a new configure switch --enable-fast-malloc, which can be either 
turned off (default, for people who want standard malloc, no matter how 
slow), or --enable-fast-malloc=full (for people like me who simply want the 
thing to be fast) and the last one is simply --enable-fast-malloc. The last 
one uses either standard libc malloc or this faster malloc, depending on 
$KDE_MALLOC (set to 1 to get fast malloc).

 Right now, it needs:
 - x86 - because of the spinlock, but it shouldn't be that hard to "steal" 
spinlocks for more platforms from LinuxThreads
 - gcc - because I'm too lazy to think about compilers not having __inline__ 
in C mode
 - glibc (only for the $KDE_MALLOC case), because it directly refers to 
symbols like __libc_malloc . In case FreeBSD etc. people are interested, 
dlsym() probably can be used instead, if their libc doesn't have such symbols 
(I should be able to test it soon on a FreeBSD machine here).

 It works for me(TM). On which platforms should I try to make it work too, 
before I commit?
 
-- 
 Lubos Lunak
 llunak@suse.cz ; l.lunak@kde.org
 http://dforce.sh.cvut.cz/~seli