[Here](https://docs.google.com/spreadsheets/d/1N6rhUfQv4gz_mwzmN5m5FKyc1xYGrzKT04Zw0LGK8gQ/edit#gid=0) are my experimental results on propagation delay.
This was discussed on bitcoin irc with @gmaxwell and @sipa.
At first, the fact that number of buckets does not matter much was questioned. You can see precise results now. 1 bucket comparing to 8 is 20-25% worse for 50% coverage, and 5-10% worse for 100% coverage.
At the same time, making R=4 itself has an impact around 80% comparing to R=2. That’s why I’m arguing for R=2 and 1 bucket (if we allow those spikes of 18 KB/s).
What is still surprising for me (and for some other people) is: with R=1 setting buckets to 1 is *slightly less significant* than doing the same with R=4. We expected it to be *more significant*, because we expected that with R=1 nature of relay through incoming should matter more.
I’ve briefly seen some graphs of @sdaftuar and it was something similar to what I’ve had — the same surprise.