Attach gdb to one of the httpd child processes and reload or continue working and wait for a crash and then look at the backtrace. Do something like this:
$ ps -ef|grep httpd
0 681 1 0 10:38pm ?? 0:00.45 /Applications/MAMP/Library/bin/httpd -k start
501 690 681 0 10:38pm ?? 0:00.02 /Applications/MAMP/Library/bin/httpd -k start
...
Now attach gdb to one of the child processes, in this case PID 690 (columns are UID, PID, PPID, ...)
$ sudo gdb
(gdb) attach 690
Attaching to process 690.
Reading symbols for shared libraries . done
Reading symbols for shared libraries ....................... done
0x9568ce29 in accept$NOCANCEL$UNIX2003 ()
(gdb) c
Continuing.
Wait for crash... then:
(gdb) backtrace
Or
(gdb) backtrace full
Should give you some clue what's going on. If you file a bug report you should include the backtrace.
If the crash is hard to reproduce it may be a good idea to configure Apache to only use one child processes for handling requests. The config is something like this:
StartServers 1
MinSpareServers 1
MaxSpareServers 1