why run_coroutine_threadsafe so slow?

Bắt đầu bằng cách chạy các ví dụ hai vòng lặp và một vòng lặp được cung cấp với uvloop 0.14.0, Python 3.6.5 và nền tảng đã nêu, sau đó so sánh thời gian xung quanh asyncio.run_coroutine_threadsafe và await my_sum(). Hoàn thành nghĩa là xác định nguồn gốc của độ trễ bổ sung giữa các thread hoặc ghi lại lý do nó được dự kiến, kèm theo một bản tái hiện hoặc benchmark tập trung.

• uvloop version: 0.14.0
• Python version: 3.6.5
• Platform: centos 6.3 cpu:Intel(R) Xeon(R) Gold 5218 CPU @ 2.30GHz

I have a need to run my code in two loop, demo code like below

import threading
import asyncio
import time
import traceback
import uvloop
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())

g_loop = asyncio.get_event_loop()
print("g_loop", id(g_loop), g_loop)

async def my_sum():
    result = 0
    for i in range(1000):
        result += 1

async def func():
    global g_loop

    try:
        while True:
            s = time.time()
#            await my_sum()
            future = asyncio.run_coroutine_threadsafe(my_sum(), g_loop)
            result = future.result(1)
            e = time.time()
            print("total", (e - s) * 1000 * 1000)
            await asyncio.sleep(0.00001)
    except Exception as e:
        print("in this======", traceback.format_exc())

def thread_loop():
    loop = asyncio.new_event_loop()
    print("local loop", id(loop))
    task = loop.create_task(func())
    loop.run_forever()

t = threading.Thread(target=thread_loop)
t.start()

g_loop.run_forever()
t.join()

then, the print time cost about 100us ~ 150us

total 107.52677917480469
total 100.13580322265625
total 154.97207641601562
total 144.48165893554688
total 141.62063598632812
total 135.18333435058594
total 174.99923706054688
total 99.89738464355469
total 174.7608184814453
total 102.04315185546875

but when I run my code in one loop

import threading
import asyncio
import time
import traceback
import uvloop
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())

g_loop = asyncio.get_event_loop()
print("g_loop", id(g_loop), g_loop)

async def my_sum():
    result = 0
    for i in range(1000):
        result += 1

async def func():
    global g_loop

    try:
        while True:
            s = time.time()
            await my_sum()
#            future = asyncio.run_coroutine_threadsafe(my_sum(), g_loop)
#            result = future.result(1)
            e = time.time()
            print("total", (e - s) * 1000 * 1000)
            await asyncio.sleep(0.00001)
    except Exception as e:
        print("in this======", traceback.format_exc())

def thread_loop():
    loop = asyncio.new_event_loop()
    print("local loop", id(loop))
    task = loop.create_task(func())
    loop.run_forever()

t = threading.Thread(target=thread_loop)
t.start()

g_loop.run_forever()
t.join()

the time cost about 40us ~ 80us

total 40.0543212890625
total 39.577484130859375
total 39.577484130859375
total 39.577484130859375
total 76.2939453125
total 40.0543212890625
total 39.10064697265625
total 81.53915405273438
total 40.0543212890625

Does change a thread need extra 60us ~ 70us? why so slow？