I did, but that won't work for gmaxwell's theoretical attack will it?

The "birthday problem" calculates the odds of ANY TWO addresses in a set matching EACH OTHER.

Unless I'm misunderstanding gmaxwell's post, he's talking about ONE address in a set (the work) matching a specific given address (the contract).

If I roll a 20-sided die until ANY number shows up more than once, we are looking at the "birthday problem".
(Odds improve with every roll until eventually odds are effectively 100%)

If I roll a 20-sided die once, and then roll it again until that initial number re-appears, then we are looking at gmaxwell's post.
(Odds remain at 5% on every roll, with an average of one success out of 20 rolls)

Am I missing something here?

In the "birthday problem" presented in the OP, a collision may occur after 2⁸³ attempts, but the odds of that collision being with an address that actually has bitcoins in it (or ever will) is extremely unlikely (since at any given time there won't be more than 2⁵¹ addresses associated with a UTXO, and in reality significantly less).

In gmaxwell's post, I don't see how finding a collision that doesn't collide with the contract address helps?