One needs to find a correct hash to claim a block. Once you have done this work you need to do it again, again in order for your block to become the longest and for other miners to consider you branch to be the correct one. If you have half the hash power of the entire network you would expect to get every other block (over time) but, with half the hashpower. it is not inconceivable for you to get 5 blocks in a row (1:32) or even 10 in a row (1:1024).
Today, 2017, it's virtually impossible for one entity (not talking about a mining pool here) to get half the network hashrate. Still - there are scenarios where 51% attacks are possible.
My claim is stronger: with probability 1 a 51% attacker will be successful (as I see it, it follows from Satoshi's paper). What about the ordering question: What is the attacker's gain of changing the ordering of transactions?