RE: Voice over IP - performance

I'm assuming in the case of, e.g., a 2650 + dual T-1 PRI interface can actually encode/decode 48 simultaneous g729a voice streams without issues? Any idea what the CPU utilisation is - or is this handled in separate DSPs in the voice network module itself?

On these particular Cisco boxes, the DSP does the all audio filtering, CODEC functions, echo cancellation, jitter buffering & adjustment, silence suppression (AKA voice activity detection, if you turned it on), and also prepends the RTP and IP headers. The router CPU just has to forward the packet that's generated by the DSP.

Router CPU utilization is therefore a function of the number of packets per second that the voice card generates and the size of each packet, plus signaling overhead. The packet size and rate depend on the CODEC itself (higher compression CODECs generate smaller packets), the sample size (20ms is the Cisco default, reducing or increasing it makes the packets smaller or larger and the packet rate higher or lower, respectively), and whether voice activity detection is on (roughly halves the packet rate).

If you leave the default settings in place (no VAD, 20ms sample size), you'll be OK with any of the CODECs.