In case anyone is victimized by this situation, this got resolved by supoprt. Apparently despite what certain VSphere engineers or VMWare zealots in your own company may say about single core vm's perhaps being preferable for vsphere cluster performance, MQM appears to _really_ need 2 CPU cores in a virtual server (and the install guide doesn't put that info in quite the right place for you to notice this). As soon as the server was fed a 2nd CPU core, the problem vanished never to be seen again (even when dropped back to 1 core). We're running with 2 CPU's now.
Also, check available disk space for ... say, the possibility of the server group not allocating all the virtual disk space to the actual C:\ volume. MQM of course won't complain in any overt way about it, but it certainly couldn't be helping matters.
Finally, and possibly most importantly given that an RPC process was the thing going bananas here, make sure you have your firewall flows set up correctly between MQM and MEG. The network port requirements for MQM are NOT well spelled out in the MQM guide at this time. Make sure your firewall flows for tcp/80 and tcp/49500 are BIdirectional. Despite what you may hear from certain SE's or even consultants, MQM will want to reach out to your MEG's (spam is stored on the MEG's and MQM sends comms to get it released, I'm led to believe) and MEG's will need to initiate connections inbound to the MQM's as well (as you'd expect), so make sure the firewall flows are bi-directional. The MQM code won't tell you there's a problem in any straightforward way, and the product guide doesn't list the port requirements succintly (yet), but they're definitely important. I have a suspicion these were the root cause of these issues whilst all the other adjunct issues surely didn't help.