Last night benchmarked on cc2.8xlarge EC2 instance ( 2 x Intel Xeon E5-2670 processors) got ~35 h/s

Thats 16 physical cores, 32 threads

same here

I restarted my computer and got 587 now as well. It was issued at 16:07 so that was a very unlucky hour for all the miners.

looks like this hour is gone be any better:)

looks like it's block per hour now

Yes, it looks like that.
I don't know how the difficulty adjustment on this coin is, but it looks as if some miners have stopped and left difficulty high for the miners that are left.
We will see if the difficulty will drop in the next hours. I'm adding another 4-core machine later today but that will not make the difference I guess

I nuked the instance after an hour... but.. its currently ~$0.253 hour on spot.

note 8 hour is the maximum time taken to try out all possibilities for a single block. A block could be found sooner, other users could find a block in meantime needing a restart...

Taking your 8 hour estimate, for EC2 mining to be successful, a single block needs to be worth > $2.024  at current difficulty.

Another thing i noticed was while mining, the memory usage for bitcoind was swinging wildly from couple of megs to > 1 GB . Is it possible to allocate memory once and re-use that? I think a lot of time may be wasted in memory allocation and subsequent garbage collection.

Is there some documentation about the hashing algorithm used? (Sorry im not fluent in reading c++ code)

Just use the addnodes that are posted in this thread.
When I use these I have no problems connecting and I get 7 or 8 connections in a few minutes.

I'd like to add this. If anybody has step-by-step instructions on how to compile this library and include on Windows - help would be gratefully accepted.