I hope this clears up the misunderstanding.
mos is much slower than bi-polar (BJT) and I will tell you why. If any of you are an EE or CE and took VLSI you should have been tought the physical structures of both of these transistors. Lets start with mos. You have a p-substrate with 2 n-wells being separted by a thin silicon oxide layer. so the current is flowing sides ways from drain to the source. the gate sits right above the oxide. upon a voltage on the gate a channel is developed along the thin oxide to complete a connection. So an easy way to think of this is by holding your hand up flat and horizontal in front of you (this being the channel). The current will actually flow from from the sides of your hand... mean having the smallest amount of contact surface area. now this is where BJT differs from mos. Bjts current actually flows vertically from the top of your hand down. Thus, a much larger surface area for current to flow. this allows the transistor to turn on and off quickly therefor you can run at much higher clock rate. (study VLSI)
Now you may ask why did they give up on the bi-polar idea and picked up the cmos. Well for many reasons CMOS transistors can be designed much smaller than BJTs. there can be at least 4 cmos transistors that could fit in the area of a BJT and they just continue to find ways to make cmos transistors smaller. Second reason is the power dissipation issues. Because mos has a much higher impedeance they dissipate less heat. Because of the larger surface area of contact of BJT and more often they are on they produce much more heat.
So many of you have been noticing Cmos is not getting any faster. Why? Well it certainly not because they can't make the transistors smaller. Transistors are getting smaller but they are finding that the copper traces are not scaling well. This why research is being put into optical and carbon nanotubes. Carbon nanotubs have amazing electrical properties but they are having a problem fabricating them. Multi cores are not doing well either. The only thing that has been found to parallelize well is graphics and servers. check out amdahl's law to find more information on this.