"Blood samples had 5-6% unmapped reads compared to 7-18% in saliva. Evaluation of source of unmapped reads found that 0.54% mapped to viral species in both sample types. In contrast, 10.3% of unmapped reads from saliva samples mapped to bacterial species, compared with only 1% of unmapped reads from blood samples as expected1. This analysis represents, to our knowledge, the first comprehensive examination of WGS and WES data generated from blood compared to saliva. Our analysis shows a high level of concordance between the two sample sources, and as expected few somatic variants. This indicates that high quality sequencing data can be derived from saliva samples for germline genetic analyses." [1]

and

"SNP genotyping results for saliva derived DNA (n = 39) illustrated a 98.7% concordance when compared with blood DNA. In conclusion, when compared with blood DNA and tested on the DMET array, saliva-derived DNA provided adequate genotyping quality with a significant lower number of SNP calls. Saliva-derived DNA does perform very well if it contains greater than 31.3% human amplifiable DNA." [2]

[1] https://www.researchgate.net/publication/309849150_Saliva_is...

[2] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3309006/

Frankly I don't think they bother. It's not a huge issue.

Sequencing will run to a certain coverage (or depth), aligning multiple fragments so that each nucleotide is sampled e.g. 30 times, which would weed out contaminants. Saliva is generally considered 'good enough' (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3497576/)

They might just use an anti-microbial agent in their collection kit to prevent growth of bacteria until it reaches their lab (plus you don't want bacterial nucleases to fragment your DNA).