Improving qPCR assay detection limits

[Hriss]

Hi guys,
I am trying to design a real-time PCR assay for copy number determination of an antibody gene in mammalian cells, using TaqMan fluorescent probes.
The issue is, I need to find a way to improve the sensitivity of the assay. So far, it works perfectly in the range 10^8 – 10^4 copies of either antibody chain per reaction (acceptable efficiency, Pearson correlation, amplification is every three cycles). When 10-10^3 standards are added to the standard curve, however, they all cross the threshold line at about 24 cycles along with the 10^4 standard, irrespectively of the fact there is far less DNA in them, and they should cross the line a lot later.
So, far, we have tried optimising primer concentrations and cycling conditions, tried different primers, ran the assay on a different machine…. The problem still exists.
Any bright ideas or suggestions on what to try next?

Thanks,
Hriss