Separation means- Lexical analysis phase takes care of tokenization and parsing phase takes care of syntax checking. i.e., the two tasks of tokenization and parsing are separated - makes software maintenance and development easy- in object oriented model we get 2 objects here, one for lexical analysis and one for parsing.
In option A, it says that if we don't use a separate lex phase, our CFG will be much bigger and this can increase the parsing time. CYK algorithm for parsing runs in $\Theta \left(n^3 . |G|\right)$, where $n$ is the length of the parsing string and $|G|$ is the size of the grammar in CNF form. https://en.wikipedia.org/wiki/CYK_algorithm