gistfile1.txt

MongoDB benchmarks

Total impression count for each page of aufeminin
224.394172 s

Unique stickers count for each page of aufeminin
230.136377

Count of impressions by day
222.475197

Apparently request time grows linearly with the number of docs in the collection
The only way to improve performance, apart from indexes, seems to be sharding between several machines.
I we want to keep our request times reasonable, we should be able to reduce them by a factor of 10 at least.
With sharding, this would involve using 10 machines which is not affordable on mongolab.
WIth 700 more mb of logs per day, using sharding to reduce request times means adding new machines every month. 
This isn't a convenient solution.