The bug has been opened on https://launchpad.net/ubuntu/+source/gnome-panel/+bug/87584

"Binary package hint: metacity

All applications launched from gnome-panel launchers, menu, or alt-f2 share the same pgid. Gnome developers claim this is not typical behavior for the panel, so may be a result of an Ubuntu customization. The same applications launched from the terminal do NOT share PGID. This is a problem for several reasons (some highlighted in this forum thread: http://www.ubuntuforums.org/showthread.php?t=347023):

1. Pressing Ctrl+C in Matlab with shared PGID closes gnome-panel.
2. fbpanel cannot raise applications via the taskbar with the shared PGID (it can raise applications fine if they don't share PGID). The applications receive focus, but do not raise. This may also tie into metacity, as it may not be a problem with other window managers, but again I think it is Ubuntu-specific. Note that the windows properly receive focus, they just don't raise.
3. Python's IDLE environment fails to raise the shell window when that window receives focus (e.g., by pressing F5) after the initial opening. This is only true if running metacity, e.g., xfwm4 properly raises that window each time it receives focus.

I submit this as a single bug report, because it appears to be an issue in how the customized gnome build Ubuntu provides handles raising windows and sending focus when applications share a pgid (which is really a metacity problem, but it is compounded by the odd gnome-panel behavior). This is true at least in release 6.10.
...
To clarify - it is the group ID (PGID) that is the same, not the PID. A simple "ps -ej" in the terminal shows each process with its own PID, everything launched from the gnome enviornment sharing a PGID but other programs with different ones, and most everything sharing the SID.

Gnome - when this was submitted as a feature request to the gnome team, they responded that they had no code that would specifically cause this (which is perhaps not quite the same thing as saying they aren't responsible). Of course, I have no idea if Canonical has added any code that would explicitly do this, either.

I have a somewhat customized environment, but it is more or less pure gnome with routine apps and a few extras (except for a couple of java apps and vmware, nothing outside the repositories, and while I am running an extra xfce4 panel for my taskbar in addition to gnome-panel, the window manager/etc is unchanged). However, as the thread discussion shows, this isn't just my environment doing this - I'm more or less just trying to summarize the discussion on that thread and elsewhere, plus some of my own research.

Also, possibly relevant update - when I compiled the newest version of fbpanel (4.5 vs. 4.1 in the repositories), it seemed to interact with metacity more or less as I would expect. So perhaps this all ties into some relatively recent in metacity? I'm just guessing."