Actually what is the relationship between number of samples on each
pool and the coverage yield?
This depends on the amount of individual libraries you put in the pool. You will be making individual libraries for every sample. Quantitating them individually. Mixing them to make a pool (with variable amount of samples, so as to balance them, in theory) and then quantitate the pool again before loading it on the sequencer. While it is possible to get reasonably balanced number of reads (with practice), there is no guarantee that read numbers for each sample will be even (they never are).
We have sequenced thousands of samples in a pool so it is feasible to do so as long as you have enough unique index combinations available to label your samples.