From `tracert 124.150.152.209`, it looks like I have 5 real sections of the 24-node trip:
* Home to Comcast (13ms)
* Comcast to cogentco.com in Dallas (5ms)
* Within cogentco.com (50ms, deterministically bouncing around US cities)
* cogentco.com US to cogentco.com Tokyo (100ms)
* To/within SE (5ms)
It's no surprise that the trip across the ocean is the heavy hitter, but randomly losing 50ms bouncing around the US is a bit weird. Apparently tracert can be a bit unreliable with these providers because they deprioritize responding to ICMP traffic, but /shrug. It matches my observed UDP response times per wireshark and squinting.
To answer my own question, my understanding from googling it is that it's just an enterprise ISP effect. When you use a VPN, you're obviously using their ISP. They are surely paying for a enterprise ISP service, so instead of your normal ISP routing, you get:
* Whatever normal ISP routing to the VPN server
* Your VPN's Enterprise ISP routing to the exit point
* Local routing from the exit point to your destination
So it may make it faster, but it may make it slower depending on whether your local ISP sucks and how far out of the way your VPN's server is. I have to imagine that for me it would decrease ping, but 50ms in FFXI isn't really worth the hassle.