Right, but it might not be the case that the low priority task is taking “longer than expected”, but just that it was interrupted in the middle of its operation (perhaps by the middle priority task).
The “inversion” being talked about is that the unrelated middle-priority task has effectively interrupted the high-priority tasks ability to run, by blocked the (related by resource) low-priority task. By giving the low-priority task (temporarily) the same priority as the higher-priority task wanting the resource that is has, it is given the CPU resource at the high priority so it can complete and turn the Mutex over to the high-priority task.