Interesting thought, let us know if you have any success. If I recall correctly the Minion has a fairly long minimum length to read (I forget if it's few hundred minimum or if it's as large as 1000 bases as minimum length) but either way you will need your bacterial mRNA to be quite intact; sometimes bacterial mRNA comes out quite degraded, and you may find some organism-specific bias to your ability to get intact mRNA. Also depending on complexity of your community, the error rate may be unacceptable for a read counting type of application, e.g. few of your reads (with 10% error rate) will correspond exactly to any of the organisms in your community, so how easily will you be able to call genes or assign reads to a particular community member?

I don't see why it should not work. You can do RNA-Seq with PacBio. The advantage would be obviously that long reads would contain several genes. Reads can be over 50kb long. But (1) the consensus quality is just 95% (at the moment, depending on the coverage - you could correct using icorn2 and Illumina reads) (2) the coverage is in general not very high (like in Illumina runs) - so very abundant transcripts (and I think probably also shorter) get far more coverage, so your representation might be very biased. I would rather do the genomic long read sequencing and if needed RNA-Seq with Illumina.