@piinecone: Interesting! I didn’t try it, but it seems that it should resolve the issue, as long as the flag is set in the editor: it has its intended value from the beginning and the bodies would be initialized properly in InitBody(). But if one sets it in code instead of in the editor, then the problem probably still persists.
In general, I feel that this flag is handled in a way that is unnecessarily prone to errors: the flag has both transient effects during every tick and non-transient effects during initialization. At the same time, users are permitted (and lead to, by the preceding comment block?) to change it directly, thus getting the former but subtly missing the latter.
@oricohen: While you are at it, maybe you could do something to protect others a bit from stepping into this pitfall? Maybe just add a warning to the preceding comment block or something, in case that you are not planning to fully refactor this thing?
ps. I played around with it some more and just realized that in fact, bEnablePhysicsOnDedicatedServer does not even disable physics for the skeleton on the physx level! I’ll post another bug report about that in a moment.