I think I finally have all the pieces together.
1) Steam by default has a setting which automatically detects when you speak and transmits your voice (To disable it: Go to Steam > Settings > Voice tab and select "Use a push-to-talk key to transmit voice". and then restart Steam)
2) Windows by default lowers volume of non-communication sound by 50% when it detects communication activity (see JKarthik answer how to change this behavior)
3) Metro apps completely ignore this setting when they are at the foreground as they're considered "immersive", but once they're in the background, they are no longer considered "immersive"...
Every one of these setting sounds reasonable by itself, but together they actually create quite a disastrous user experience...