Taking over someone else's code, I've discovered a bottleneck. This is a portion of a script that accepts a BAM and outputs smaller, sorted, rmdup-ed, indexed BAMs for each chromosome.

for chr in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 X Y M; do

    samtools view -@ $threads -b $bam $chr > $bam.$chr.bam
    samtools index $bam.$chr.bam
    samtools rmdup $bam.$chr.bam $bam.$chr.rmdup.bam 
    samtools index $bam.$chr.rmdup.bam

done

After reading this Biostars question, I replaced the loop and the samtools view with

bamtools split -in $bam -reference

Though it can't be multithreaded, this does seem—at least in my benchmarking so far—to be faster. I'm wondering, though, whether the BAMs it produces are guaranteed to be sorted and rmdup-ed if the input $bam was? Or would I still need to run samtools rmdup (or samtools sort) on each child BAM?