If you use calibration curves to quantify your samples relative to an external standard, then the efficiency does actually not matter. If you do not use external standards, like many people do when they perform a "relative quantification" where they more or less directly compare ct-values, differences in efficiencies between your gene of interest and your calibrator gene (loading control/reference gene/housekeeping gene) will render your conclusions wrong (imagine: you start with identical amounts for both genes, getting identical signals; the model expects equal ct-values. If the efficiencies are not equal, you will in fact get quite different ct-values, so you would conclude that the concentrations were verx different; note that this is a simplified view; the thing is more complicated in real "relative quantification" experiments where the differences between target and calibrator gene are compared between different samples - but the errors introduced by differening efficiencied do no cancel out, so the problem remains).

A general point why "low" efficiencies may be critical is as follows:

In qPCR you have idel reaction conditions, a vast overshoot of primers, nucleotides, and enzyme. Why should not all molecules in a sample be amplified? Why only, say, 80%, and why should this be consistently the case in different samples and over the time course of the PCR? If I use a standard prep to measure a calibration curve giving a suboptimal efficiency, how can I be sure that some other sample will amplify with exactly this same efficiency?

Since we lack a theoretical understanding of the biochemical causes of a sumoptimal apparent efficiency, there might be other reasons leading to a wrong estimate of this efficiency, possibly in the way we pre-process the fluorescence intensity data (I think here especially of background- and trend-corrections and of normalizations). All these calculations are based on heuristics, and in some cases these data processings may result in amplification curves that do not well represent the essential biochemical process we model to get the quantification.

But if the quality of data is in question, the results are in question, too.

At least for "close-to-ideal" efficiencies the results can/may be explained based on our understanding of the process and it at least seems reasonable, so that we may have reasonable hopes that the measures and results we get truefully reflect biological properties (rather than data-processing artefacts).

If you use calibration curves to quantify your samples relative to an external standard, then the efficiency does actually not matter. If you do not use external standards, like many people do when they perform a "relative quantification" where they more or less directly compare ct-values, differences in efficiencies between your gene of interest and your calibrator gene (loading control/reference gene/housekeeping gene) will render your conclusions wrong (imagine: you start with identical amounts for both genes, getting identical signals; the model expects equal ct-values. If the efficiencies are not equal, you will in fact get quite different ct-values, so you would conclude that the concentrations were verx different; note that this is a simplified view; the thing is more complicated in real "relative quantification" experiments where the differences between target and calibrator gene are compared between different samples - but the errors introduced by differening efficiencied do no cancel out, so the problem remains).

A general point why "low" efficiencies may be critical is as follows:

In qPCR you have idel reaction conditions, a vast overshoot of primers, nucleotides, and enzyme. Why should not all molecules in a sample be amplified? Why only, say, 80%, and why should this be consistently the case in different samples and over the time course of the PCR? If I use a standard prep to measure a calibration curve giving a suboptimal efficiency, how can I be sure that some other sample will amplify with exactly this same efficiency?

Since we lack a theoretical understanding of the biochemical causes of a sumoptimal apparent efficiency, there might be other reasons leading to a wrong estimate of this efficiency, possibly in the way we pre-process the fluorescence intensity data (I think here especially of background- and trend-corrections and of normalizations). All these calculations are based on heuristics, and in some cases these data processings may result in amplification curves that do not well represent the essential biochemical process we model to get the quantification.

But if the quality of data is in question, the results are in question, too.

At least for "close-to-ideal" efficiencies the results can/may be explained based on our understanding of the process and it at least seems reasonable, so that we may have reasonable hopes that the measures and results we get truefully reflect biological properties (rather than data-processing artefacts).

the responses are in

http://link.springer.com/article/10.1007%2Fs12223-013-0255-5#page-1

Jochen is right. The efficiency shouldn't affect the accuracy of your quantitation. Some PCR's are just not all that efficient, but the standard curve takes care of all that. I've operated PCRs with efficiencies around 65% and still got good results.

However, low efficiency does affect the sensitivity. Increasing the number of cycles helps, but eventually amplification of nonspecific products and primer dimers will outcompete your target,

Also, if your efficiency suddenly drops, there's something funny going on and you should troubleshoot it. I always keep a small sample of QC-approved reagents in reserve to check this.