bwa mem does not find all alignments found by bowtie in multi-hit mode

[aliner]

I generated a set of simulated reads from one microbial reference genome and would like to find all alignments of these reads against a large set of genomes. I ran

bwa mem -a

using an combined index generated from a set of genomes, which included the genome from which i simulated my reads (let's call it A) and many other genomes. The SAM output had alignments for all reads, but for some reads there were no alignments against A, which was not the case for

bowtie2 -a

or even BLAST.

In narrowing down the problem, I found that if I made a BWA index using A alone (no other genomes), I got alignments for all reads against A. This was also true if the index had A + 9 other genomes, and A + 99 other genomes. But with A + 999 other genomes, some of the reads were not aligned to A though they were aligned to other genomes. I have tried various command line parameters to lower the alignment "stringency", even using

bwa mem -a -k10 -d 10000 -c 1000000 -B 0 -O 0 -E 0 -L 0

, but to no avail.

I was under the impression that the -a parameter would give me all alignments of a certain quality against all genomes in the index. bowtie2 appears to do so, but not BWA. Does anyone know why? Am I missing some command line parameter? Or perhaps a new branch of code? Or is this a property of BWA?

Thanks!