All WPT MSE framesize tests are known to be flaky (having occasional failures) and have been so for a very long time. https://bugs.webkit.org/show_bug.cgi?id=220103 improved the handling of video size but wasn't enough to fix these tests. The failures are probably caused by rapid flushing as the tests append overlapping ranges. Flushes received before playback has started are tricky to implement since GStreamer elements are generally not well tested for that use case.
The test or tests filed under this bug are not failing anymore. Test expectations updated in https://commits.webkit.org/258724@main.