For starters, you can use address.send(x) instead of address.call.value(x)(). All computations on Ethereum have to be funded with "gas" (transaction fees), and send() only forwards a small amount of gas. The recipient can write to the log but that's about it. TheDAO used .value() which forwards all the available gas.
If you do use .value(), then you can do it safely by doing it only once per method, as the very last step, and only using it in top-level methods. Then any sort of reentrant callback will find any required state changes already done. E.g. subtract from the user's balance, send the funds, and if the send fails then throw. The recipient can call back but the balance is already decremented. Using a mutex is another option. I've tested all this and it works.
There are a lot of reasons Ethereum is designed this way. The most common reason for a contract to run code when receiving funds is to prevent users from accidentally losing their money. E.g. in any contract that holds ether and maintains a ledger with balances for multiple users, if anyone sends ether to the contract address, they'll lose it, unless the contract either throws an exception or automatically adds to the user's balance on its ledger.
User addresses and contract addresses are interchangeable, because it lets you do things like use a multisig timelocked vault to receive funds and anything can send to it just as if it were a regular user account.
All that said, the devs are working on improvements too.