With good technique, you can reliably perform RT-PCR or RT-qPCR on a sample size of about 100 cells. Some of the comments are correct that you will need to compensate for the reduction of the initial cell input with additional PCR cycles. However, each magnitude in reduction of cells corresponds to about 3 extra PCR cycles (2^3 = 8 ~= 10), so your overall qPCR cycle number will not change significantly.
Since you are dealing with a small amount of cells, and isolating by TRIzol, the pellet could be difficult to see. I would suggest you use something like Glycogen (Glycoblue from Ambion) to co-precipitate the pellet into something more visible. You would add this prior to the Isopropanol to precipitate your nucleic acids.
Your use of extra TRIzol is not a problem unless you're lab is short of money. It will dilute your concentration of nucleic acid, which will make precipitation harder, but using glycogen will compensate for this. Using an amount less than the manual directed ratios can be problematic as it could result in incomplete lysis and denaturation of your cellular components.
As user137 recommends, Pre-amplification of the products is not a good idea as it will likely distort your final results. I'm not sure what gene expressions you're looking at, but it is always a good idea to include multiple housekeeping genes as baselines to calculate your relative expression numbers, provided you have enough RNA isolate to use as input.