Just to follow up on my previous post, the more I read the IEEE papers, the more unsure I am about the answer. In papers like "Tail Current Noise Suppression in RF CMOS VCOs", JSSC Vol 37 nos 3, current source is identified as the main culprit for noise while in papers like the one in [Hajimiri JSSC 01 Vol36], tail current noise source is very small. Maybe it is just the difference in optimization approach that yield such a great difference?
(one report noise summary @ 3MHz the other @ 600khz so that partly explain the different. But still, there are many literature that claims that the tail current source dominates the noise at 1/f^3 and 1/f^2 region and the data in Hajimiri paper seems to suggest otherwise....)