More compression, worse latency. More bandwidth, worse latency.
"You only need 100Kbps for voice and that's for full encapsulation using G.703 with a SIP overhead. With more efficient codecs you can drop that to 10Kbps or lower."
Show me how. Include figures for needed bandwidth and resulting latency.
Codecs might get the voice bandwidth down to less than 10kbit/s, and on a circuit switched setup (e.g. GSM) that works quite well.
Now try doing that in a packet switched network where the IP overhead bandwidth per conversation is much biglyer than the emcoded voice bandwidth per conversation.
Number of conversations per Mbit/s and latency are worse (from an end user point of view) than you get even from ISDN.