Windows undeterministic test timeout

I have got a very long standing bug in one of my Rust projects that I really have to fix - it is about time.
TravisCI and Appveyor are configured for the project and since this commit I experience random (undeterministic) timeouts for the "newly" introduced integration tests.
The funny thing is that at TravisCI all tests always work, then and now.

Appveyor was configured for stable, beta, nightly and x64 and i386 and one or more integration and/or doc tests is always on timeout on a random configuration.

I really cannot think about anything that is heavily influenced by the underlying operating system.
Maybe you can point me to some interesting things that I haven't thought about already.

Edit:
For clarity, here are the CI reports:

I have analyzed the test-timeout failures in the recent history of Prohpet on appveyor. Throughout all build targets parameterized with (stable, beta, nightly), (MSVC, GNU) and (x86, i386) failure distribution seems pretty much completely random.

Seems like I have to dig much further into this and probably setup some Virtual OS on my computer to not start a CI-bughunt.

Tips, help and hints are still very welcome!

EDIT

I have adjusted the appveyor settings so that debug output is shown on the console, clearly showing that the algorithm (neural network) does not properly approximate the valid area. This is strange since it is so random and only occures on windows/appveyor builds.

EDIT 2

One test that fails particularilty often has its source here.