Hello,
Finally I had time to check real values in following scenario.
Two 3CX servers (7.1.6589.0) connected by two Master/Slave UDP bridges in local LAN. Bridge was set to use G729 codec.
Server #1 only one not registered dumb extension and Digital Receptionist playing on-demand Dead Can Dance 'Martha Song'. No users connected except of remote desktop session.
Server #2 with two registered extensions making one and two simultaneous connections over the bridge to the server #1 digital receptionist.
Bandwidth measurements - Paessler's PRTG Monitor reading SNMP bandwidth counter from the Cisco Catalyst switch server #1 port.
There is visible difference between one and two simultaneous calls. Because the source of the RTP stream was internal (server #1 IVR) only background usual windows domain 'noise' and remote desktop session affects observations (and server performance in generating RTP stream, however quality was good).
It looks like bridge overhead is minimal, there is nothing like bandwidth reservation, and there are no measurable differences in traffic with media server up or down. Only visible with wireshark proofs of bridge activity with no calls transported are SIP registrations dialogs.
Any comments?
Jacek