Here, researchers at ecSeq Bioinformatics provide detailed performance comparisons of NGS read aligners. They stress that benchmarks only measure specific aspects and may not be used to claim any universal superiority or inferiority of a particular tool.
In order to compare different short read aligners, the researchers use a published real-life paired-end DNA/RNA-Seq dataset. All optimal alignments (also multiple mapping loci) of 100,000 read pairs of each sample were obtained by RazerS 3 (full sensitivity mapping tool). In the benchmark shown below, they measured the performance in finding all optimal hits of different NGS mappers with default parameters. True positives are reads with up to 10 multiple mapping loci, allowing up to 10 errors (mismatches and indels).
Note that they explicitely want to find all multiple mapping loci in this benchmark and not only unique mapping loci or just one random hit of several. The researchers believe that reads mapping multiple times should not be discarded since gene duplications and repeat regions are known to be biologically relevant.
Visit the benchmark page for more information.