The charts listed under "Performance Trend Per Test Case" view show the values from the last sample instead of averages for that URI. This can be confusing when the build is unstable because some URI was slower, but the numbers listed in the build console output don't match the ones shown on the trend per test case charts. The same URI on the same build can have a widely different value on the chart, as it's not the average, but the last sample.

The bug is caused by TestSuiteReportDetail::doRespondingTimeGraphPerTestCaseMode adding all samples to the dataSetBuilderAverage structure. I suspect it was intended to average those, but in fact as the column and the row keys are the same for all the samples, it will just overwrite the previous sample value and the last one is used for displaying on the map.