There are probably many reasons. One may be the weight of inertia. If you know a technology and it works for you, there's little reason to switch to something else. For some applications, microarrays may be cheaper than NGS, especially if you factor in the cost of analysis because data handling and analysis is also more complex for NGS than for microarray. In addition rapid changes in technology and pipelines makes NGS feel like a less mature approach than microarray.
CGH arrays are employed to detect chromosomal aberrations. Focussing on human genetic disorders, most of the major hospitals and diagnostic centres in the US have happily adapted microarrays. Thousands of samples are analysed on daily basis for a number of patients. Microarrays are allowing analysis , QC and sample preparation in large batches with cheaper costs.
Secondly, as mentioned above, researchers are comfortable with the analysis. Companies like Illumina and Agilent are providing free-of-cost workflow based software for microarray analysis which requires minimum knowledge of computers. On the other hand, most of the NGS based tools are open source and requires basic understanding of Linux. The transition is happening gradually but it will take time because the technology is different, the raw data is different and the down-stream analysis is also very much different.
I don't have much experience with genotyping arrays but I think these are an example where arrays are still superior to NGS. If you want to profile a sample at known SNPs genomewide, then it's going to be difficult for NGS to beat arrays in terms of cost, amount of data to process and accuracy. I think with arrays you can genotype tens or hundreds of samples for the price of a whole genome sequence and you need to process gigabytes of data instead of terabytes.