I didn't do any in-depth analysis of query performance, but I did run the query for the same amount of time for the three tables, and observed the compressed tables ended up with more records in them at the end of the three min run. Not sure what (if anything) you can read out of that.

Good question on the compressed backups of compressed databases. I hadn't thought of that. Since my test database contained a combination of compressed and uncompressed tables, not sure of the answer to your question. Should be fairly easy to test though.