I am curious how and if this system deals with concurrent modification of the underlying database by other processes. I can't find anything about that in a brief glance at the paper. That doesn't mean it's not a useful technique for a class of problems, but I guess most installations are going to have more than one client. I guess some sort of cache sharding might do the trick in that case.

This is quite different from that. Read your link to see why: the mysql cache invalidated all cache entries if the underlying table changed at all, and only cached whole result sets. This program does things quite a bit differently, tracking much more than just "which table did this result set come from?"