Hello, I dont use Drosophila but Arabidopsis, however the decision flowchart is the same. Basically what we do for our project is to start with a pilot project of a few libraries of your sample and control, but here a few guidelines.

1) Although Illumina can give you amazing depth, more replicate is better than more depth (one transcript in a million is not very significant from a biological standpoint). Most people go for few reps because libraries is the expansive part but it should be the other way around.

2) Is you sampe homogenous or heterogenous ? Example: if you are working with cell culture in triplicate (homogenous) versus individuals of various ages (heterogenous) will dictate (in part) whether you need few or many replicates.

3) Is the change you are expecting big or small? If you are comparing transcriptome of Drosophila after a 24h cold shock at 4C compared to a room temp control, obviously you will not need many replicates to see a difference. In our case we want to see the effect of the expression of one deregulated gene on the whole transcriptome...many reps!

4) When you get your raw read,  proceed with QC. First analysis you do is saturation curve. It may have a different name but basically it consist of plotting the number of unigene you have with 1 rep. How many with 2 rep, how many with three rep, etc. You will have a curve that plateaus, good cost/info ratio is right after the inflexion point, if you are in the plateau you have over sequenced.

I think it is a good start. For our most recent pilot we started with 6 libraries for control and 6 for transgenic (we had 20 lines to do), those 12 sequenced in a single lane. It turned out that 4 libraries showing little variance and dispersion was generally enough so we did 80 libraries for the remaining, at 16 libraries per lane.

Hope that helps.