I am assuming by miR control you mean a stable reference (aka housekeeper) gene.

Despite raising awareness of potential pitfalls of qPCR, use of experimentally validated reference genes for data normalisation is still widely disregarded.

The use of small nuclear RNAs including RNU6 and RNU1A as reference genes still remains popular even without experimental assessment of stability. Furthermore, RNU6 and RNU1A are small nuclear RNAs and thus do not accurately represent the miRNA population. In my experiences these two small nuclear RNAS were unstable as well.

To do this analysis you will need to assess the stability of >3-4 miRNA in your samples. Search the literature for miRNAs expressed in your tissue/model.

Then you will need to test whether these miRNAs are stable in your experimental groups using a normalization program such as GeNORM/Bestkeeper. 

"I'm using RNU6-B but I don't get the same Ct values on my samples, even if they are cultured samples."

I am assuming you use the same RNA and cDNA inputs across your samples. Sybr green assays have a limitation when it comes to distinguishing minimal sequence similarity. Perhaps there is template-specificity issue here. How is your melt curves for this? Any primer dimers?

Have a read of this paper

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2327352/

Indeed, find a stable reference gene is a hard work, especially in injury conditions such as cancer, diabetes, etc.

The ideal is test some miRNAs or "small RNAs" (at least 3) to chose the most stable in all your experimental groups. You can test the stability by some specific programs as "geNorm, Normfinder, BestKeeper...".

I do not see any problem in use small nuclear or nucleolar RNA as reference gene. You can use too miRNAs stable, if you find one.

You need "surfing in the literature"...

Although not with cancer, this is an interesting paper with reference genes for miRNA and mRNA:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4623139/

Good luck!