Depends how much hardware continues to remain on most machines. If the market eventually shifts towards dumber and dumber boxes (as people keep predicting) that centralised model will still be sweet for performing things like encoding.
Remote encoding of VoIP doesn't make much sense, especially if the V is video. Processing power will be cheaper than bandwidth for a long time when you're talking about this much data (CD quality mono audio: 700kbits/s, 720p30 4:2:0 video: 332mbits/s).
