After considerable thought and discussions with other people, I affirm that there is no error in Rosenfeld's Bancor paper as Dan Larimer suggests.
Therefore, there seems to be no theoretical explanation (so far) for the individual formulas implemented in convert_to_exchange and convert_from_exchange at exchange_state.cpp (EOSIO Dawn 4.0).
However the composition of both formulas, lead to the conversion between EOS and RAM in such a way that the product of the connectors reserves (EOS times RAM) is constant. This means exact agreement with Rosenfeld's paper (https://goo.gl/HXQBUr) section "multiple reserve currencies" taking S=S_0, m=2 and F_1=F_2.
Thank you @bytemaster for your answers, attention, everything.