I am a novice in Ehrenfest dynamics and would like to know why Ehrenfest dynamics suffers from the "overcoherence problem"?
Generally speaking, Ehrenfest dynamics tries to model nonadiabatic processes. These are processes where an electronic wavepacket propagates on several potential energy surfaces. The different surfaces have different forces associated to them. This means that the parts of the wavepacket on the different surfaces move in different directions. Ultimately, this leads to decoherence - the parts on the different surfaces do not know about each other anymore.
Ehrenfest dynamics produces only one trajectory that runs on all the potential surfaces in lockstep. This produces an artifical superposition of excited states, which has no possibility for decoherence.
You can start with the discussion around Figure 3 in J. Chem. Theory Comput. 2019, 15, 5031 and work your way up from there.