If you're really, really interested in absolute quantification, then it makes sense to include a standard curve on each plate, but I certainly wouldn't do so routinely.

Primer efficiency is worth determining, but this is also a physical property that reflects the sequence, the buffer conditions and the cycling conditions (annealing temp etc), and should not really change between runs assuming you're using the same conditions and buffers. So yeah: determine efficiency once, establish that it somewhere between 95 and 105% efficient, and then don't worry about doing that again.

(This has nothing to do with reference genes, either: your reference and your GOI primer efficiency are two entirely separate things)

I also wouldn't necessarily advise doing this with actual cDNA, since "what effect does cDNA dilution have" is actually a different question -the buffer concentration changes with each dilution, after all, and what you're really measuring is when this buffer is no longer inhibitory to your reaction.

Ideally, you should establish the threshold beneath which cDNA synthesis buffer is no longer inhibitory to your reaction, and then just...use that dilution, for everything. Or just go with 1/10 or 1/20, because both of those will be fine (I use 1/20 but use 2ul per well because I find it easier to pipette slightly larger vols, but 1/10 and 1ul per well would be effectively the same thing).

For efficiency, I would usually run the qPCR once, run the product on a gel and make sure there was one obvious band (melt curve should support this). Then I'd cut that band out, purify it, calculate the number of molecules.ul-1 based on amplicon length and concentration, and then use that to make my std curve.

This tells you efficiency, and also LLOQ - the template amount below which your assay is no longer truly quantitative. Usually this is somewhere between 1-20 molecules, because this is stochastic partitioning territory: it's not easy to reliably quantify targets at such low levels because "10 molecules per microlitre" really means "10 per ul on average": you could get 5 molecules in one well, 10 in the next, 13 in the third, giving you wild Cq variance (you can address this by using more replicates, but it's usually not really worthwhile).

For relative quantification, if the primers and samples are the same, I reckon that a standard curve rendered once should suffice.

qPCR primer standard curves should be done to validate the primers prior to use. then those conditions are good for those primers given:

1. you use at the same final concentration as the validation reaction

2. you use the same cycle as the validation reaction

3. you use the same SYBR/TaqMan master mix as the validation reaction

4. you use the same machine as the validation reaction

if you change any of these then you should re-run your standard curve to recheck efficiency.

you do not need to run a standard curve on each plate, UNLESS you are trying to do an absolute quantification of target rather than relative quantification.

to get appropriate comparable data with least variation all experimental samples should be processed as a batch for cDNA synthesis, then make a master mix for qPCR to test all samples and run them all on ONE plate. if they do not fit on one plate then you need a plate control: which would be an identical sample run on each plate in the analysis. I usually use my control treatment sample for this and repeat it on the extra plates. if the Cts for the plates differ for the plate control then you normalize to the plate control first, then to your housekeeping genes, then the control group.