I've been a little bit busy those days (or it's better say: months!) but I'm catching up with Npgsql bugs little by little.
Today I fixed a bug which was making Npgsql to not return the correct number of rows affected by an update when using DataAdapter.Update() method. This bug was floating around for a long time! Thanks Christian for the heads up and testcase!
Some weeks ago, I played with another bug which was causing the famous "stream doesn't support seek operations" exception. It turned out that that the large object api isn't handling errors correctly which is letting data on the stream which causes the problem. I have an ugly workaround for this which is simply flush the stream before trying to write to it again. I sent a message to Npgsql dev list explaining this issue with more details.
That's it for while. I'll be hunting down the remaining bugs and will post progress here.