Right, in the sequencing step the Simulator limits the number of reads to maximally the number of molecules in the cDNA libary, respectively twice this number in the case of paired-end reads. The reason behind this is that in silico sequencing would otherwise oversample the library and artificially many reads come to fall at the same position.
Therefore, in order to have a sufficiently big repository of cDNA molecules you have to start with enough molecules in the expression simulation step. The number of expressed RNA molecules does not always translate 1:1 to the number of molecules subsequently derived cDNA library, the relation is strongly dependent on the settings of the applied protocol, respectively. However, the correlation $\frac{number RNA molecules}{number cDNA molecues}$ is about constant for little or large datasets.
To this end, I recommend to perform a little run, with few million initial RNA molecules and the corresponding library construction parameters, to estimate the ratio between RNA and cDNA molecules. Pedro, a friend, found with his settings for 5 million initial RNA molecules 2.81 million cDNA fragments, estimated factor is 0.562. In order to produce about 10 million reads, he estimated consequently to start with an initial RNA population of size ~17.8 million. From this run he derived 9.8 million cDNA fragments.
Best,
Emmett