A small handful of people can still easily cost millions of dollars per year. And DSN time to talk to the rovers is expensive, probably in the neighborhood of $2,000 per hour.
How much DSN time do you need. Given the deley, I assume the only use is batch transfers, so cost per bit seems like the more useful metric (along with how much data is involved)
$5M a year seems like it would be very reasonable. The original 90-day price tag was $820M, and they've spent an additional ~$200M over the past decade and a half. Maybe it's really not worth it to spend $5M/year to extend the life further, but the scientific returns would have to be quite severely diminishing.