it usually happens when there is a short time (like a couple of seconds) between two blocks.
Miners can start work on block n+1, even before finishing downloading block n body and can sometimes solve that n+1 block in that time and submit it rightaway.
I suppose it could also happen when there aren't many txs in queue (as in the case these days) because most or all of the transactions picked up by the latter miner were already included in that previous block, so the miner goes ahead and submits the empty block... to save the time it takes to download and verify new txs (the time it could take for someone else to find and submit a block).
So it's generally deliberate, as everyone's competing to be the first to find the next block. It's definitely deliberate when network's heavy.
