It sounds like you have kwon issue for 4.1..x.

Do you use ob_end_*() functions? Any where in your script?
What is your 

output_buffering
and 
zlib.output_compression

ini settings?

I forgot about ob_implicit_flush ;)
Do you use it?

I use default settings:

output_buffering = Off
zlib.output_compression = Off

and don't use any output buffering functions

Then you are having crash bug(s).
First grab snapshot http://snaps.php.net and check if you still have the problem. (Make sure you build with --enable-debug)

If you still have problems, check your httpd error logs. If you see segfaults, send a backtrace. 

Unfortunately, my home OS is Win2K-Server, and I myself don't have VC++ to try to build PHP from sources...
Can I provide any useful information for you, using my OS software which I already have?
I have both IIS and Apache web servers.

If you can create _short_ & _complete_ script that reproduces the problem. We can fix it :)

I tryed to make some small scripts before, but there were no problems with them. My site is complete enough, there is a lot of includes (but execution time for homepage on my PC is less than 1 sec.). It looks like there are no problems for simply scripts, but I'll try again to make something ...
By that time, if you like you may watch the results on different servers:
1) php4.0.6: http://www.sec4.wfdns.com/www.freetakeout.com/
2) php4.1.1: http://vh110014.radntech.ca/fto/ 
- you may try to click some links there ("log-in to your account", for example) to see the difference in cutting point.

Sorry. I wanted to say "... My site is complex enough..."

Could you set up linux/freebsd box to reproduce this bug? 
And generate backtrace. (Or get exit code with gdb)

You can write "error_log(__FILE__.'(.'__LINE__.')')"
to locate to find ofending function/line also.

I made test using the advised method, and found that crush appears in one my simple function "call". A simple example goes below:

<?
// --------------------------------------\
// Apply specified function to arguments
//   if 2'nd argument is an array, its elements replace arguments
//   it's a way to apply function to arguments by their references
// Examples:
//   list($c1)=call("intval",$a1);
//   list($c2,$c3)=call("htmlspecialchars",$b2,$b3);
//   call("htmlspecialchars",array(&$b2,&$b3));
// ------------------------------------\
function call() {
  if (func_num_args()<2) return false;
  $argv=func_get_args();
  $func=array_shift($argv);
error_log(__FILE__.'('.__LINE__.')');			//
  if (is_array($argv[0])) $argv=&$argv[0];		// Crush is here
error_log(__FILE__.'('.__LINE__.')');			//
  reset($argv);
  while (list($k,$v)=each($argv)) $argv[$k]=$func($v);
  reset($argv);
  return $argv;
  }
// ------------------------------------/


call("intval",1);		// OK
call("intval",array(1));	// Crush
?>

Note: actually cutting line goes some earlier than that crush point, but, I think it's because some PHP bufferization, and is not important now.

I've found a decision of problem: I simply need replase line
if (is_array($argv[0])) $argv=&$argv[0];
to:
if (is_array($argv[0])) $argv=$argv[0];
And I'll not lose any fuctionality features.

So, crush appears when an array element assigns to the same array by reference. Althow this operation is not very often and very useful :-), it works correctly and w/o crush in 4.0.6 version, and I think this crush should be fixed though.

reclassified. You should also try the latest release 4.2.1
(just in case someone already fixed this)

It seems this bug does not appear in 4.2.1. Checked on Linux/Apache.