It's fairly hard to come up with direct suggestions regarding the APIC interrupt delays without a better understanding of the rest of the OS. 
It happens that I recently modified our virtual EFI firmware to use the APIC timer in one-shot mode instead of the PIT, and didn't notice any such problems, but the firmware environment is probably much more straightforward than most OSes: It's only uniprocessor, for a start, and makes very limited use of interrupts.
One of the coolest ways to investigate such problems is with VProbes. Check out the VMware Community for VProbes. You can use VProbes to instrument a running VM to measure and detect all sorts of interesting things, some of which might give you insight into the delayed timer interrupts.
Other than that, the only suggestions I can come up with are fairly generic and probably stuff you've already done: Try to reduce the problem space by simplifying things. Does it happen if the guest OS is running with only one vCPU? Does it happen if you disable all but the utterly essential drivers? I imagine that you've checked the LVT programming a hundred times already. Double-check that your reprogramming of the initial count register can't lead to races. Check that any higher-priority ISRs are correctly issuing a single timely EOI.
Good luck!
--
Darius