Briefly, the quantile normalization forces the entire empirical distributions 
of each array to be identical while the scale method forces only the medians of 
each array to be the same. 

The cyclic loess method, which is more sophisticated and more time consuming, 
is based on the repetition on each pair of arrays of a loess regression between 
the difference of expression between the two arrays and the average expression. 

We recommend the use of the latter which gives a less agressive and more robust 
transformation of data. 

For details, see 

Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, and Smyth, GK (2015), 
limma powers differential expression analyses for RNA-sequencing and microarray
studies. Nucleic Acids Research 43, e47.