@Peter: The RNG is very susceptible to giving similar answers for similar seeds. For example, If you use any seed less than 53668 and generate one bool, it will be True. I expect, but have not confirmed, similar behavior with seeds the size of the current unix time. Nick makes a similar point, "All RNG algorithms have biases", but trust me way I say core/Random has very noticeable biases!
If you do not have access to ports, and cannot accept a seed during initialization from someone who does, then I understand the desire to use the current time. However, the current time can only be obtained through a signal, which completely destroys the purpose of a pure random number generator.
I don't think either of these reasons alone are insurmountable, but combined, it's just a massive headache.
@d13: See Jason's excellent answer.
@Jason: Thanks for that post. Glad fissioning is helpful.
@Nick: Good point about concurrent instances... or components. And regarding the mathematics, each bit of the initialization value should have a 50/50 chance of being a 1. Timestamps (in human-sized time) don't have this property.
One way around this problem would be to have a seed initialized in a sensibly random way at program start. The seed would be immutable for the duration of the program but change with subsequent programs. There hasn't bee a whole lot of support for this idea, but maybe you all can bring it up again?