I have tried a lot of things, but whenever a single test fails, the performance publisher plugin does not create graphs for individual tests. Only the overview (sum) of all tests is created. If no test has failed, the graphs are created as expected.
Since I cannot think of any reason why someone would want this, I must assume that this is a bug.
My Jenkins job runs the tests with MSTest. Each test measures the execution time for various scenarios and each test creates a new xml file that is parsed by the performance publisher plugin. If a test reports an error (e.g. the execution time exceeds a certain limit), the job is marked as unstable. This is done automatically by the MSTest plugin.
The plugin works fine with multiple xml files, as long as all tests have passed.