Exactly this. I haven't looked at the Reader architecture myself, but take a hypothetical "Hey, let's open-source Reader" project at Google:
Reader probably uses the Googlebot crawl infrastructure to retrieve feeds. We're not going to open-source the crawler, so now we need to rip that out and replace it.
Let's say Reader used Megastore to store feed data (the Megastore paper says "hundreds of applications" use it, so this is not an unreasonable idea). Megastore is a nice chunk of intellectual property in and of itself, plus it's built on top of Bigtable.
Throw in Chubby for locking/coordination and some MapReduce jobs for bulk processing, and you're basically down to either rewriting the entire core of Reader or open-sourcing the "crown jewels" of Google's infrastructure. Not to mention any number of underlying libraries (e.g. core C++ libraries) that have hundreds of engineer-hours of investment.
Reader probably uses the Googlebot crawl infrastructure to retrieve feeds. We're not going to open-source the crawler, so now we need to rip that out and replace it.
Let's say Reader used Megastore to store feed data (the Megastore paper says "hundreds of applications" use it, so this is not an unreasonable idea). Megastore is a nice chunk of intellectual property in and of itself, plus it's built on top of Bigtable.
Throw in Chubby for locking/coordination and some MapReduce jobs for bulk processing, and you're basically down to either rewriting the entire core of Reader or open-sourcing the "crown jewels" of Google's infrastructure. Not to mention any number of underlying libraries (e.g. core C++ libraries) that have hundreds of engineer-hours of investment.