merkle root mismatch, that's weird. I have only seen that once on a ubuntu server because it had the wrong g++ installed.
it looks like the submitted blocks via "SubmitBlock" are hashed/checked wrong - at least with a different hash than the self mined blocks..
Both mined and submitted blocks are checked by function "CheckBlock", which is called by function "ProcessNewBlock". I'm trying to find out why is the merkle root different, sadly I don't have much knowledge of MPOS a Stratum daemon.