In my case, I mapped the RNA-seq reads to both host and parasite genomes simultaneously, after concatenating their genomes. Then, I removed unmapped reads and reads with primary alignment to the host chromosomes (I was interested in the parasite). From my point of view, if you remove reads that map to the host genome without verifying if they map to the parasite genome, you could end up removing reads derived from parasite's genes that have highly conserved homologs in the host genome, even though such cases are rare.

You should use bbsplit.sh from BBMap suite for this purpose. It will allow you to bin the reads to multiple genomes (and handle multi-mappers in intelligent ways). See this thread.