Indeed, generation time is one of the big things to take into account. That's why in the early days, the estimates that the genetic clocks gave was only good enough for closely related species.

But the method has come a long way. They have "relaxed clocks" now, which take into account generation time and stuff (it's all Bayesian statistics which I don't really understand). When they compare to known fossil records, it broadly agrees. But I think even now you'd expect it to be more accurate when comparing rodents together than the whole of arthropods for example.