You're looking at this from an engineer's perspective. Yeah, great, we can build Cassandra -- who cares? Google papers have tons of performance benchmarks -- how is it scientific if none of those are reproducible?
Am I defending academics who can't or won't turn over all their data? No, of course not, but most of them don't try to pretend you can do science and still be proprietary.
P.S. I'm not talking about all companies. Intel and MSR turn over a bunch of stuff. I'm talking specifically about Google.
"Hadoop was derived from Google's MapReduce and Google File System (GFS) papers."
So apparently the disclosure was good enough to give you that :)
http://www.slideshare.net/jbellis/cassandra-open-source-bigt...
Shows that cassandra is based on the bigtable paper, among others (hbase i believe was as well).
Their disclosure is apparently good enough to get folks going ...
FWIW: In reading 20 years of compiler papers, i've had more luck getting code for papers from commercial companies than academics. It's become somewhat better over time, but plenty of universities still seem completely unwilling to give code to papers.