This is an interesting attack vector.

To avoid this the idea is to have several oracles which behave exactly like multisig on the bitcoin blockchain. For example, 2 of 3 oracles which need to agree on the BTC/USD price on various bitcoin exchanges. This way the attack you were describing is harder to accomplish (but not impossible).

There is new exciting research on oracle signatures which would make signatures and verification more efficient and we could easily have NofM oracles with N>10 or more.