Well, usually TriplePlay equipments (the ones which offer data, voice and video on the same device) use separate VLANs (virtual LANs) for voice and data and possible even a third one for video.
This happens precisely for the purpose of tagging data and voice with different VLAN ID and inside the VLAN ID with different QoS (quality of service) classes.
QoS is a mechanism for prioritizing packets/frames based on their Layer 2 or Layer 3 Cos (class of service) value, inside the frame tag or in the DSCP field of the IP header. It is a very intricate subject in iteslf but the end result should be that frames/packets with higher priority (i.e. voice and video) should be given priority by your modem/router device over traffic on the data VLAN. This happens transparently and without any configuration from your part (all should be already done by your ISP when they install you the device).
Now, if you’re just using VoIP calls over a simple data VLAN like it seems to be your case (as in Skype or messenger calls over an Internet connection which provides data services) all your traffic goes on the same VLAN as the rest of the data so you don’t benefit from any traffic classification from your router.
In this case your VoIP traffic doesn’t have any “emergency lane” to take so all you can do is to free up as much bandwidth as you can and hope for the best, as kluelos suggests.
@ kluelos: Actually, I think that by suspending the active tasks BC accomplishes two things not just one:
It stops all outgoing traffic, which frees up the uplink immediately;
It either sends a TCP ACK with window size 0 on each TCP connection (which effectively will grind to a halt the incomming traffic on each connection from any remote host trasmitting towards the local client) or a FIN/RST packet which will close the TCP connection. I haven’t actually sniffed BC traffic to analyze it and see if what I mention at the second point happens for sure, but I’ve empirically seen that actually the download bandwidth is becoming virtually instantly available for other applications upon suspending active connections, which suggests such behavior. Besides, it would be plainly stupid to just drop incomming traffic **after **it passes and burdens your whole LAN up to your PC, without notifying the remote hosts not to bother sending anymore when the means to do it, exist.
I don’t argue that the former peers of the swarms for your running torrents may still try to connect to you for about the half hour you mention, but those will merely be TCP SYN packets which are very small and which won’t find any echo on your side, so there won’t be any real bothering traffic going on since there are no connections being set up. The only traffic that you really won’t be able to do anything about will be the UDP traffic, but in terms of bandwidth that’s usually negligible on today’s connections.