I have anecdotal evidence that posting tasks through the v8 platform is slow (slower than the usual post task through TaskScheduler from Chromium that is).

Here's a proxy to infer this:

V8.GC.ParallelTaskLatencyMicroSeconds is always slower than TaskScheduler.TaskLatencyMicroseconds.Renderer.UserVisibleTaskPriority (which is a superset of V8.GC.ParallelTaskLatencyMicroSeconds).

https://uma.googleplex.com/histograms?endDate=20180226&dayCount=1&histograms=TaskScheduler.TaskLatencyMicroseconds.Renderer.UserVisibleTaskPriority%2CV8.GC.ParallelTaskLatencyMicroSeconds&fixupData=true&showMax=true&filters=platform%2Ceq%2CA%2Cchannel%2Ceq%2C1%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

Appears to be true on all platforms.

I think this might contribute to the reason why we see the main thread sometimes being inactive for a while when kicking off parallel GC jobs (ref. ItemParallelJob). "idle" as in no trace events for a while until the main thread kicks off its task (I'm adding task_scheduler.flow tracing category to help highlight this for all PostTasks that land in TaskScheduler).

I'll try to save a trace next time I see this.