Imagine a toy scenario where a cancer will kill 80% of people with it at exactly 3 years, and all 20% remaining will uh live full lives (so say another 20 years).
If you created a drug that shifted the ratio to 60/40, and measured right at the 5 year mark. In a population of 100 untreated people, you'd get like 4080 total months lived (80x36 + 20x60). In a population of 100 treated people, you'd get like 4560 months lived (60x36+40x60).
That's only generating an extra 4.8 months of life per person treated. However, you've doubled the number of people alive at the end of trial.
Now obviously this is a toy example that's probably a bit exaggerated, but this type of behavior is exactly why median/average life extension is an inadequate measure alone.