I think with 100 runs of same code on same data I can see all exception of protobuf-net !! about protobuf-net HOT 5 OPEN

Do you have a repro that actually demonstrates this? Is it possibly that your 
your rig uses concurrency and you are simply using the same thread/data from 
parallel threads?

I don't know when I'm gonna see an exception next; but will update this issue 
with details if any raised. 

About concurrency: both from mine and BPlusTree's side, it's a single-threaded 
process. 

I also tried Tags as well; as it suggested by @Marc for some questions on 
stackoverflow, but no improvements for my conditions ! 

@Marc, as you have already checked BPlusTree's source, I'm adding the piece of 
code I use for update, "just in case" 
KeyValueUpdate<C, B<C, M>> updateFunction = delegate(C Key, B<C, M> Value)
{
   B<C, M> block = Value;
   block.lambda.Add(new Lambda<C, M>(tau, interval.metadata));
   return block;
};
di3.TryUpdate(key, updateFunction);

What I would really love to see is a complete runnable example where I don't 
need to guess at lots of details, so that I can actually try to repro...

hmmm ... OK either I should summarize the code (which then you may not see the 
error :D). OR: I can share full project with you, but to run it, somehow I have 
to pass you ~10GB ENCODE data to be able to execute. What do you suggest ?

BTW, the issue is just raised again (after 3 successful runs) 
https://www.dropbox.com/s/v7t2n1gll3guc0e/exception.txt?dl=0

I've an idea: updating serialized messages happens quite often in my scenario 
(<20M/50M ), and disabling update everything "seems" to be working fine. This 
encourages me that my assumption with dynamic messages (i.e., 
http://stackoverflow.com/questions/26223017/length-prefix-with-dynamic-messages-
in-protobuf-net) is true ... do you agree ?