I've found nice Technet blog which gives some advices when to use which replica init method, but without any technical detail. To make a good decision I would like to know how the replication using Use an existing virtual machine as the initial copy actually works. There is no requirement for snaphosts, not changed VHDs, just the virtual machine ID has to be the same. Does it read whole disks on both nodes, checks block hashes and sync blocks which differs?
Yes, that's exactly what it does.