It would be very interesting if this data were broken down both by more than just income, e.g. ethnicity [1] and immigration status. Unfortunately, the effect being described could easily be caused solely by composition changes.

A hypothesis I'd love to see tested: immigrants have shorter life expectancies [2] and are more likely to be in the bottom 50%. Therefore, an increased proportion of immigrants reduces the average life expectancy of the bottom 50%.

[1] Ethnic gaps in life expectancy are huge. Asians life 10 years more than blacks, for example.

[2] Might be tough to get data on this. My understanding is that immigrant life expectancy data is weak because many immigrants go back home to die (which skews the numbers).