One way to record selected handsets across a global system would be to install a localised copy of Echo at each of the locations and mirror the handsets you need to record to the second network card of the local Echo server.
Call playback will be done from the localised Echo server to either a local agent PC, or a remote agent PC, only when requested so the only additional bandwith across the WAN will happen while calls are being listened to.
Yes you can! The port(s) on the network switch where the SIP data leaves the office would need to be mirrored to the second network card on the Echo server.
Some hosted providers don't offer Call Detail Records either so you can use the ones that Echo produces instead.