Funnily enough, 1.18.03 was the last version to do this right... weird.

Interestingly, the latest Schism version from Storlek's Bitbucket also fails two tests that Schism didn't fail back in 2009 when the tests were first administered. We've fixed those now, except for something that could be described as the four-dimensional equivalent of a corner case (the first note of test 25, which OpenMPT also plays incompatibly but isn't even worth reporting).

I know about test case 25, it isn't really worth fixing at this point since it means rearranging a lot of code (which would then be handling other formats incorrectly).
Player routines constantly change so it's no wonder if some supposedly fixed tests suddenly fail again. I want to introduce a regression framework to automatically catch such errors in OpenMPT, but it's not there yet.

Anyway, fixed in r6362.