This problem does not occur on 4.3.0, with the same configure line (any basic configuration will do).  Sadly, I cannot provide any real error information, as it is not crashing, but the behavior is fairly predictable.

I have a open_basedir setting in php.ini, confirmed via phpinfo(), its value is not important.  On each of the configured virtual hosts I have a different value using php_admin_value.  However, on some virtual hosts I turn safe_mode off using php_admin_flag and do not set an open_basedir value, so it should be using the php.ini setting.  Turning safe mode off is probably not relevant, but trying to provide any related information.

When requesting pages on that virtual host (the one with no overriding open_basedir setting, and safe mode off), the requests will randomly fail.  When checking the error log, it is a normal exit and php is logging that 'open_basedir restriction in effect ... '

The path that it provides as the active open_basedir value is not correct.  It in fact randomly changes to other virtual hosts settings.  It does not always fail, it may take 20 requests before it does (even a simple auto-refreshing page, will fail shortly with that error).

I believe there may be a bug in reverting to the default configuration value, and using an old value in memory.  I am using prefork in apache, so it would most likely have to be an uninitialized pointer to the same memory space, memory spaces for new forked processes should be repeatable.  

Well thats about all I can provide, it can be fixed by explicitly stating the open_basedir value in each virtual host (so ones where before I kept it at the default value, now explicitly set the default value).

I'm more than happy to follow any instructions to get further information on this, just not sure what to do as its not an actual crash or anything.