We track the heap insertion count (a simple counter, which may overflow, each time an entry is inserted) as a means to ensure that timers with the same firing time are ordered by insertion in the heap. Currently this count is a process-wide (i.e. globally unique across all threads). But it doesn't need to be and it increases the rate at which the count can overflow (since multiple threads can increment it). Architecturally it makes more sense for this count to move to per-thread storage.