This is a real old problem, and I guess if noone want to capture all traffic until this happens and does some real deep analysis, this maybe will stay for even longer. But if I get you right, this not happened with the old method. I'm pretty sure, as the repo should point to the same code, this shouldn't make a difference, but maybe you could run a short diff to verify this. Are there any other things different?
Do you compile both with you're checkintall workflow? You've installed the other variant of the asio code, as it's easier to iinclude that one with cmake, so maybe this makes the difference or in general the socket could be the factor that influences this.