Actually, slow in this case does not at all mean "this operation isn't implemented in hardware or optimized assembly".

The slowness of an algorithm like bcrypt is a tunable property of the key expansion algorithm. A step in this process will be repeated 2^n times, where n can be configured by the user.

If a password function was only slow because it wasn't implemented in assembly on your server, an attacker would obviously just go implement it in assembly for his brute-force crack.