I've not tried very hard. In short kmer-based genotyping does work (see Mykrobe.com and
nature.com/ncomms/2015/151221/ncomms10063/full/ncomms10063.html) but Cortex variant discovery is not likely to work well because I don't think the graph cleaning will work very well. I've only done limited analysis of this. If you put the reads through Gene Myers' scrubber thing and removed chunks of reads that are clearly bad, then you might get good results ...
regards
Zam
PS Sorry for slow response.