First and foremost: YES, I KNOW! This is completely senseless microbenchmarking and micro-optimization. I ONLY LOOK AT THIS FOR FUN.

That being said. The 64bit multiplication is faster, and also constant in time. I've measured this using the timestamp counter which doesn't necessarily count actual CPU clock cycles, but possibly an integer multiple of it (seems to output always numbers divisable by 8 for me).

    (...)
    100 loop:01100100(152) 64bit:01100100(48)
    101 loop:01100101(128) 64bit:01100101(48)
    102 loop:01100110(160) 64bit:01100110(48)
    103 loop:01100111(128) 64bit:01100111(48)
    104 loop:01101000(192) 64bit:01101000(48)
    105 loop:01101001(88) 64bit:01101001(48)
    106 loop:01101010(152) 64bit:01101010(48)
    107 loop:01101011(128) 64bit:01101011(48)
    108 loop:01101100(160) 64bit:01101100(48)
    (...)

That's on an Intel(R) Core(TM) i5 CPU M 450 @ 2.40GHz and the cycles include the rdtsc, a few movs, and the call to the respective function.

https://github.com/vogelchr/bitstring_via_64bit_mult