Just tried it (on Ubuntu, apache 2.2.20), and it worked fine for me, even without fiddling with the default_* / master_* options.   I.e., the ssh master connection etc. is automatically being terminated properly.

Maybe enabling debugging

$Net::OpenSSH::debug |= 0x7f;
[download]

will shed some light on what isn't working in your case (output ends up in the webserver's error log).

Thanks for the suggestions. The error "Pseudo-terminal will not be allocated because stdin is not a terminal." seems to be common. Below is the relevant part of the error log. The problem occurs frequently but not every single time the page is loaded.

# open_ex: ['/usr/bin/ssh','-S','/var/lib/wwwrun/.libnet-openssh-perl/
+root-192.168.100.8-20273-555951','-l','root','192.168.100.8','--']
Pseudo-terminal will not be allocated because stdin is not a terminal.
# _waitpid(20296) => pid: 20296, rc:
# open_ex: ['/usr/bin/ssh','-O','check','-T','-S','/var/lib/wwwrun/.li
+bnet-openssh-perl/root-192.168.100.8-20273-373748','-l','root','192.1
+68.100.8','--']
# waiting for slave, timeout: 5, remaining: 5, sleep: 1
# _waitpid(20299) => pid: 20299, rc: Interrupted system call
# open_ex: ['/usr/bin/ssh','-S','/var/lib/wwwrun/.libnet-openssh-perl/
+root-192.168.100.8-20273-373748','-l','root','192.168.100.8','--']
Pseudo-terminal will not be allocated because stdin is not a terminal.
# _waitpid(20300) => pid: 20300, rc:
# open_ex: ['/usr/bin/ssh','-O','exit','-T','-S','/var/lib/wwwrun/.lib
+net-openssh-perl/root-192.168.100.8-20273-373748','-l','root','192.16
+8.100.8','--']
# waiting for slave, timeout: 5, remaining: 5, sleep: 1
# _waitpid(20301) => pid: 20301, rc: Interrupted system call
# open_ex: ['/usr/bin/ssh','-O','exit','-T','-S','/var/lib/wwwrun/.lib
+net-openssh-perl/root-192.168.100.8-20273-555951','-l','root','192.16
+8.100.8','--']
# waiting for slave, timeout: 5, remaining: 5, sleep: 1
# _waitpid(20302) => pid: 20302, rc: Interrupted system call
[download]

Looks like your system isn't automatically resuming interrupted system calls.¹

You could try modifying the module's code in order to do it manually.  Resuming a system call manually means to simply call it again if $! is EINTR or ERESTART.  In this particular case, look for line 755 in OpenSSH.pm

    $r = waitpid($pid, WNOHANG) and last;
[download]

and modify it to read

    do {
        $r = waitpid($pid, WNOHANG);
    } while $r<0 and ($! == Errno::EINTR or $! == Errno::ERESTART);
    last if $r;
[download]

This should take care of this particular waitpid call which produced the message in the debug log.  There are more, though, which would likely need a similar treatment, so it's probably best to get in touch with the module's author... (he frequents this place regularly, so he'll probably see the thread sooner or later, anyway).

___

¹ System calls may be interrupted for a variety of reasons. The more loaded a system, the more often it happens.  Most systems these days automatically resume them, so code isn't always written to handle the rare case where you have to do it manually...