It really depends on what your goal is. If you are using the probe to quantify the amount of DNA/RNA present in your sample, then it is necessary to run a standard curve on each plate alongside your samples. It provides a definite reference point for your unknown samples. For example, you can determine viral copy number using that way. Typically people use 4-6 concentrations (at 1:10 serial dilution ratios) run in duplicate or triplicate.

If you are doing genotyping or presence/absence experiments, then a standard curve is not really necessary. You may want to run one plate with a curve to ensure that your reactions are working correctly and efficiently. After you have validated your assay, you don't need to include a curve in subsequent plates.

If you are doing gene expression experiments, it's better to include one or more additional "housekeeping" genes (such as GAPDH, ACTB, RPP30) against which you can compare your unknown samples.

Thank you for your detailed answer. Perhaps I omitted some information. I am planning to use the double delta method for relative quantification. I was just wondering whether even in this case when using TaqMan probes it is necessary to do a standar curve to assess primer efficiency.

If you're doing relative quantification, you can probably skip it. I often do. Typically I validate new TaqMan-style probe assays by first running the primers only with SYBR Green to get a melt curve (to determine if there is off-target amplification, excessive dimers, etc) using multiple known positive and negative controls. After that, I run the assay including the probes with the controls. If the results look good, I proceed with the assay. I haven't had any problems with this approach.