But the ratio of cost to transaction value stays the same regardless of what the value of bitcoin is, since the cost of pulling it off increases at the same rate as the transaction value does, as the value of bitcoin increases.
Don't know exactly what you mean here, maybe you are agreeing that all values are proportional to bitcoin value?
Also, your claim seems to assume that all the transaction value in a block can be confiscated by the person doing the double spend attack, when in reality, the only thing they can steal is the money they transferred to others, by reversing those transactions, and NOT the entire transaction value.
No, in my example I used only one transaction of 100$, which would not be conspicuous. And it would probably not attract any attention if a few more of those transactions were slipped in.
It would definitely be much easier to attack bitcoin if transaction volume doesn't increase by the time coin generation becomes negligible, but your calculation of cost doesn't take into account the long term investment required to acquire a large amount of hashing power, not just purchasing the hardware, but getting the facility, setting it up, etc all of which have a huge fixed cost, and therefore the need to double spend for many blocks, to make back the cost of the investment.
True, I did not take into account the initial acquisition value (which I would estimate at 10.000$ in the example above). I only took into account the depreciation of that hardware (and then added another $ per hour). I think that's not unreasonable and standard procedure in profit/loss calculations. The same trick could be pulled of multiple times with that very same hardware, or the hardware can be used for different purposes (e.g. video rendering) afterwards.