I had problems with the 0.88, 0.88.1, 0.88.2 and 0.88.4 compile on linux (and probably versions before that).

After I would compile my clamav from source
(using the ./configure, make, make install)
and put into service a new version of clamd, I would log errors immediately
ERROR: pthread_create failed
and clamd would hose itself

My qmail system with qmail-scanner would barf.

Some threads in mailing lists have suggested, somewhat tersely, to raise softlimits assuming that DJB's tcpserver were being used.
I'm using a RedHat-based system (not using the .RPM for clamd) and have elected to use the initscripts version to launch clamd.

Irrespective of my configuration, I see a few folks still struggling with the compiled clamd.  On my system, 'cat /proc/cpuinfo' yields:

processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 3.00GHz
stepping        : 9


So I took the advise of a CFLAGS optimization page from Gentoo (after some hasty googling) and did

'make clean'
then
./configure CFLAGS="-march=pentium4 -O2 -pipe -fomit-frame-pointer"

Without changing my system 'ulimits', nor my initscript, nor fussing with softlimits in a tcpserver run file, I now have clamd running without complaint.

YOUR CFLAGS will be different for your architecture, of course, but I presume something can be gained by either i) optimizing for your specific architecture, or using the '-pipe' or '-fomit-frame-pointer' bits.  I don't know compiler behavior well enough to grasp out "why", but it works.  That's enough sometimes...

Godspeed.