Im not sure if this is related to libpcap or not, but in virtual machines (e.g. esxi) the timestamps captured when sending a train of packets from A to B are much more closer in time than the timestamps measured in the hypervisor. Any document/reference that talks about this effect?
That's probably a function of the way the hypervisor works. The timestamps you get from libpcap are applied to the packets by the OS, not by libpcap (except on HP-UX, but I suspect you're not running on HP-UX, as I don't think VMware ESXi runs on PA-RISC or Itanium).
Administrators of the "libpcap" SourceForge project have superseded this tracker item (formerly artifact 3576423, now bug 178) with issue 182 of the "libpcap" GitHub project.