what resistances make nmos faster than the cmos
Sorry to comment, but I think none of you answered his question.
Maybe none of you knows why BJT is faster than MOS, although many of you tried, but your understanding is not even close.
In general, when comparing a monolithic BJT and a monolithic UJT such as MOS:
BJT has a base, intended for hole replacement. This is like a minority carrier buffer for electrons. Under the high electric field strength at the Collector, most of electrons are accelerated. Such acceleration depends on Vce and HFE.
MOS has no buffer. MOS depends on inversion (regardless weak or strong) to conduct between source and drain, thus the channel poses a considerable resistance (Ron).
As a device operates over a longer period of time, heat causes to Ron increases, this reduces maximum bandwidth.
Parasitic caps on BJT is relatively less significant than in MOS because such caps mainly between the nodes to emitter. There parasitic caps pose little limitations to BJT. However, parasitic caps in MOS exhibit influence inside the device lateral structure, inter-referencing source, gate and drain. Some are ignorable at high-frequency model, but still the inherent Cgs, Cgd are Cds are forever there!
However, MOS has evolved from long-channel to short-channel, to HEMT, FinFet and even to the extend of using SOI. The gap is closing.