is it feasible to process 8M reads with pre.cluster

I am following MiSeq SOP. I have ~8M reads in each group covering the ~1500 bp of 16S rRNA. So this 8M reads from one group can only be processed by one core. I found it took 7.5 hr to process 50000 reads. If the time required is proportional to the number of reads, it takes at least 1200hrs to process 8M reads, which makes this step unrealistic. Could you please suggest a solution? Maybe I have to skip this step? Or I can break 8M to 100 parts and process them in parallel, which could be better than skip this step?

Can you split the reads by sample group? Have you run the sequences through unique.seqs? If you can split them then you can use multiple processors to precluster the samples separately.