Performance timings on my system:

The code

import time
import enum

class Color(enum.Enum):
    RED = "Red"
    BLUE = "Blue"
    GREEN = "Green"
RED, BLUE, GREEN = Color

class FastColor:
    RED = Color.RED
    BLUE = Color.BLUE
    GREEN = Color.GREEN

def f():
    for _ in range(10000000):
        Color.RED
        Color.BLUE
        Color.GREEN

def g():
    for _ in range(10000000):
        FastColor.RED
        FastColor.BLUE
        FastColor.GREEN

def h():
    for _ in range(10000000):
        RED
        BLUE
        GREEN

import timeit
print(round(timeit.timeit('f()', number=1, globals=globals()), 2), end='  ')
print(round(timeit.timeit('g()', number=1, globals=globals()), 2), end='  ')
print(round(timeit.timeit('h()', number=1, globals=globals()), 2), end=' \n\n')

version	enum	"fast" enum	global enum
3.9	2.23	0.78	0.41
3.10	2.68	0.91	0.47
3.11	2.29	0.72	0.21
3.12	2.21	0.62	0.14

The 3.12 timings are with every enum member being accessed via the enum.property descriptor (i.e. no enum members being stored directly in the enum class __dict__).

Out of curiosity I ran the above code with member.member access, both with and without the deprecation warning:

warning	no warning
20.41	4.16

So updating code to be correct will keep performance to 3.11 levels (or better).

Uh oh!