I think I still disagree. I haven't completely thought it through, but even if you discount perfectly, it still depends on what exactly you discount BY. If you discount by EXACTLY the average rate of T-bills then perhaps yes. But why would anyone discount a prospering company by only roughly inflation plus a few basis points (where T-bills typically trade)? You would surely discount it by a higher rate than that because it is growing by more than a few real basis points.

I think what you're missing is the difference between increases in intrinsic value versus forward-looking expected returns. If you go from:
[a bundle of operating companies plus investments plus substantial excess cash]

to

[a bundle of operating companies plus investments plus far less excess cash]

you have changed the risk of the investment, and you should apply a higher discount rate to the expected cash flows of the (riskier) post-buyback enterprise.

when you say this:
After that, they don't cancel each other out anymore. That's because when you use $5B of cash to buy $5B of shares, you are disposing of something that is growing at 3.9% (nominal) a year, and buying something that is growing at something higher than that, maybe 10% (nominal) a year. In essence, you are selling a $5B T-bill yielding 3.9%, and buying(very roughly) 20% T-bills, 50% varied company stocks, and 30% wholly owned companies (or whatever the ratio is today).

I think you end up with the conclusion that replacing all your low-risk, low-expected return assets with higher-risk, higher-expected return assets is value creating. It's not true for repurchases in the same way that it's not true of buying $10 billion worth of Google for $10 billion in cash (assuming Google is fairly valued) - it's just choosing a more aggressive point on the risk/return spectrum.