问题似乎已解决。
Ted Lyngmo提到了一个非常重要的事实：
...当您包含<execution>时，它会检查是否可以找到tbb信头。如果它们可用，它将包括它们并使用tbb作为后端。如果它找不到可使用的后端，它将回退到std::execution::seq...
这对我来说是一个惊喜，如果我不显式地使用TBB头在我的代码-我仍然需要包括和链接TBB...
因此，我必须相应地修复CMakeLists.txt，以包含和链接TBB头文件和库。

cmake_minimum_required(VERSION 3.14.0)
project(EXEC VERSION 0.0.1)

set(CMAKE_C_STANDARD 17)
set(CMAKE_CXX_STANDARD 20)
set(CMAKE_CXX_STANDARD_REQUIRED ON)

# ! Set the TBB library path
if(CMAKE_CXX_COMPILER_ID MATCHES "GNU")
    set(TBB_ROOT_LIB "L:/oneTBB/mingw64/mingw64/bin" CACHE PATH "Path to TBB") # ? Change this path according to your oneTBB location
    set(TBB_ROOT_INC "L:/oneTBB/mingw64/mingw64/include" CACHE PATH "Path to TBB") # ? Change this path according to your oneTBB location
elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
    set(TBB_ROOT_LIB "L:/oneapi-tbb-2021.9.0-win/oneapi-tbb-2021.9.0/lib/intel64/vc14" CACHE PATH "Path to TBB") # ? Change this path according to your oneTBB location
    set(TBB_ROOT_INC "L:/oneapi-tbb-2021.9.0-win/oneapi-tbb-2021.9.0/include" CACHE PATH "Path to TBB") # ? Change this path according to your oneTBB location
endif()

add_executable(
    executionTests
    targets/executionTests.cpp
)

# ! When you include `<execution>` GCC checks if it can find the tbb headers.
target_include_directories(
    executionTests
    PUBLIC ${TBB_ROOT_INC}
)

# ! Also GCC need to link TBB library to use it's features, so specify path
target_link_directories(
    executionTests
    PRIVATE "${TBB_ROOT_LIB}"
)

# ! Libraries for MSVC and GCC are different
if(CMAKE_CXX_COMPILER_ID MATCHES "GNU")
    # * I found only one version of TBB dll for mingw64,
    # * not sure if's Debug or Release version,
    # * but results very close to MSVC Release
    set(TBB_LIB "-llibtbb12")
elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
    if(${CMAKE_BUILD_TYPE} STREQUAL "Debug")
        set(TBB_LIB "tbb_debug.lib")
    else()
        set(TBB_LIB "tbb.lib")
    endif()
endif()

# ! Finally link library
target_link_libraries(
    executionTests
    PRIVATE ${TBB_LIB}
)

if(CMAKE_CXX_COMPILER_ID MATCHES "GNU")
    target_compile_options(
        executionTests
        PRIVATE
        -O3
    )
elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
    # ? I never mentioned why I have these lines:
    # ? /O2 (improve performance by reducing execution time and optimizing code size)
    # ? and /RTC1 (perform runtime checks) flags can't be combined
    # ? so this is a workaround to remove /RTC1 from resulted command line
    STRING(REGEX REPLACE "/RTC(su|[1su])" "" CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG}")
    STRING(REGEX REPLACE "/RTC(su|[1su])" "" CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG}")
    target_compile_options(
        executionTests
        PRIVATE
        /O2
    )
endif()

字符串
需要提到的是，MSVC需要在构建脚本中看到配置类型，而不仅仅是在配置脚本中，所以还需要添加以下内容：

cmake --build build --config Debug --target executionTests -j 8 -v
# cmake --build build --config Release --target executionTests -j 8 -v

型
已完成有用的代码更改：
C++头文件而不是C头文件（stddef.h、stdio.h>），以及其他有用的更改：

#include <cstddef>
#include <cstdio>

#include <ranges> // ! add this for std::ranges::generate

// * I came up with this template function to easy use std::mt19937
// * and generate doubles in a functor way  
template <typename _Type, _Type _left, _Type _right>
_Type generateRandomNumber()
{
    /*
    ! Your program looks portable as-is so it's just a matter of making
    ! the implementation use a backend for the execution
    ! policies. I would however not use `random_device` as a source since
    ! that will produce different results every time. Use
    ! a deterministic source so that you get the same content of the containers every time.
    ! `std::mt19937 rd;` would be better for that.
    */
    std::random_device rd;
    std::mt19937 gen(rd());
    std::uniform_real_distribution<_Type> dist(_left, _right);    //* get random number from 1 to 1000
    return dist(gen);
}

int main()
{  
    // ! %llu is the wrong printf conversion specifier for size_t. Use %zu.
    printf("Testing with %zu doubles...\n", testSize);
    vector<double> doubles(testSize);
    // * my idea how we can generate random doubles
    std::ranges::generate(doubles, generateRandomNumber<double, -100.0, 100.0>);

    // time how long it takes to sort them:
    for (size_t i = 0; i < iterationCount; ++i)
    {
        vector<double> sorted(doubles);
        /*
        ! Never use high_resolution_clock./
        ! The specification for that clock was messed up and
        ! even the author of that clock says to not use it.
        ! Use std::chrono::steady_clock instead.
        */
        const auto startTime = steady_clock::now();
        sort(sorted.begin(), sorted.end());
        const auto endTime = steady_clock::now();
        print_results("Serial STL", sorted, startTime, endTime);
    }

    for (size_t i = 0; i < iterationCount; ++i)
    {
        vector<double> sorted(doubles);
        const auto startTime = steady_clock::now();
        // same sort call as above, but with par_unseq:
        std::sort(std::execution::par, sorted.begin(), sorted.end());
        const auto endTime = steady_clock::now();
        // in our output, note that these are the parallel results:
        print_results("Parallel STL", sorted, startTime, endTime);
    }
}

型
现在GCC和MSVC的结果有很大的不同。

调试：

• 通用条款：*

Testing with 1000000 doubles...
Serial STL: Lowest: -99.9999 Highest: 100 Time: 86.243000 ms
Serial STL: Lowest: -99.9999 Highest: 100 Time: 83.652300 ms
Serial STL: Lowest: -99.9999 Highest: 100 Time: 85.125400 ms
Serial STL: Lowest: -99.9999 Highest: 100 Time: 86.877800 ms
Serial STL: Lowest: -99.9999 Highest: 100 Time: 85.984300 ms
Parallel STL: Lowest: -99.9999 Highest: 100 Time: 35.638500 ms
Parallel STL: Lowest: -99.9999 Highest: 100 Time: 30.529100 ms
Parallel STL: Lowest: -99.9999 Highest: 100 Time: 35.590700 ms
Parallel STL: Lowest: -99.9999 Highest: 100 Time: 29.676400 ms
Parallel STL: Lowest: -99.9999 Highest: 100 Time: 33.012500 ms

型

• MSVC：*

Testing with 1000000 doubles...
Serial STL: Lowest: -99.9999 Highest: 99.9999 Time: 277.651300 ms
Serial STL: Lowest: -99.9999 Highest: 99.9999 Time: 281.134900 ms
Serial STL: Lowest: -99.9999 Highest: 99.9999 Time: 278.242000 ms
Serial STL: Lowest: -99.9999 Highest: 99.9999 Time: 280.372500 ms
Serial STL: Lowest: -99.9999 Highest: 99.9999 Time: 275.779000 ms
Parallel STL: Lowest: -99.9999 Highest: 99.9999 Time: 98.904400 ms
Parallel STL: Lowest: -99.9999 Highest: 99.9999 Time: 94.853300 ms
Parallel STL: Lowest: -99.9999 Highest: 99.9999 Time: 100.861400 ms
Parallel STL: Lowest: -99.9999 Highest: 99.9999 Time: 92.364000 ms
Parallel STL: Lowest: -99.9999 Highest: 99.9999 Time: 102.100400 ms

型

发布日期：

• 通用条款：*

Testing with 1000000 doubles...
Serial STL: Lowest: -99.9998 Highest: 100 Time: 77.569600 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 83.123500 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 81.983300 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 82.967000 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 82.845600 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 34.475000 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 34.092200 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 30.292100 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 33.041200 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 30.095900 ms

型

• MSVC：*

Testing with 1000000 doubles...
Serial STL: Lowest: -99.9998 Highest: 100 Time: 95.452600 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 98.047800 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 97.359000 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 96.975000 ms
Serial STL: Lowest: -99.9998 Highest: 100 Time: 98.612100 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 35.154200 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 35.384300 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 35.499900 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 34.143500 ms
Parallel STL: Lowest: -99.9998 Highest: 100 Time: 33.570500 ms

型
GCC又出问题了）可能我需要链接另一个调试版本的TBB库。
不过总体的问题现在解决了。非常感谢Ted Lyngmo。