You have NOT installed two different versions of Vicidial. Both installations are VERSION: 2.14-738a. That's the Vicidial version and is the same on both systems. You used two different Installation .isos, but those merely account for the operating system and Environment in which Vicidial runs. All Vicibox installers install the Latest version of Vicidial, but each will have a slightly different environment including the Operating System, Asterisk version, MySQL version and lots of other package difference possibilities. In the end, however: Vicidial is the software you interact with in the Web interface and the software that controls the calls.
Next up: You're missing a major piece of information. The hardware upon which this installation has occurred. For instance: Is this a Virtual machine? If so ... that's your problem. Install it on a physical server and you'll get your speed. Vicidial uses hi-res perl timing which requires every "tick" of the system clock. Sharing that system clock in a virtualized environment is NOT supported.
If the system is not virtualized at all, then you'll need to find out where your bottleneck is. Is the command to transfer the call sent immediately but only acted upon later? Or is the command sent after the delay?
The methodology here is that there is a perl script running and controlling the call after it is answered. At the moment of answer, extension 8368 is fired (is that immediate or delayed?). After 8368 fires, it starts up a perl script and that perl script is expected to make a decision about which agent should get the call. So ... does that script fire immediately when expected or is it delayed? Is that script running quickly (as in under a second) or is it delayed?
OR are you even using 8368 at all? Are you using a different campaign routing extension that incurs a delay such as the AMD extension?
Happy Hunting!
Oh: I see you edited it with AMD. Yep, AMD requires at least two second to run. Vicidial itself has about a 1 second delay (perhaps 1.5) so a 4 second delay is completely normal if you use AMD. Always has been. Don't use it if you want to avoid the delay. Let an Agent decide if there's a machine, and if there is have the agent push the VM button (which can be automated to leave a VM message by the system if you like, releasing the agent for the next call immediately).