Sana, I performed serial dilutions of cDNA for each of my primer sets, this gives you a standard curve that performs several functions:
1) Identifies the optimal cDNA dilution (aim to get a suitable Ct with all primers)
2) Enables you to calculate primer efficiency (not everyone uses this in their calculations but it's a good idea)
3) Provides you with a standard curve which you can use to determine a suitable threshold, if you wish.
I stopped using delta-delta-Ct and started using the REST program as I prefer it, horses for courses though, there's no one right way. If you're only using a single HKG it's not such an issue.
In terms of assessing the cause of poor reproducibility, have you checked your machine? I've found significant differences across plates, especially between the inner and outer wells on older machines using a solid heating block. If you want to assess this, make up a mastermix and cDNA solution sufficient for the entire plate, pipette it out and see what the well-to-well variation is like. Also, repeating any given experiment will give you a figure for the inter-assay deviation which is the other concern.