The secret is in a separate bare-bones software which is watching the second blockchain, in the manner that Mike Hearn described a hypothetical piece of software watching Google.
Mike was describing a case where a system watching google is trusted to perform faithfully, or is a member of a collection of such identified semi-trusted systems where it is trusted that no more than some threshold will behave dishonestly. This is normally what we're talking about when we talk about oracle mediated contracts.
Moreover, in the longer description I mention possible obfuscation techniques such as using the hash of a block as a source of randomness.
It is a hotly debated subject in theoretical cryptography if practically secure obfuscation is possible even in theory even assuming many things which would make the effort insecure, impractical, or simply useless. It is not currently _practically_ possible in any case.
I hope you don't feel I'm wasting anyone's time. If you can explain how:
a) an Oracle can sign a transaction upon Mike Hearn winning a gold medal as reported by Google...
...is fundamentally and unalterably different from...
b) an oracle can scan a blockchain and sign transactions embedded within it after certain criteria are met
...then I'll close this specific request.
There isn't but your question was unrelated to these points as far as I can tell. There are actually secure ways to achieve the latter other than oracles, however, at least in theory (google coinwitness) but practice is another matter. (And, amusingly, not the former, because SSL doesn't provide non-repudiation)