If I were writing a high-concurrency multi-threaded system to handle millions of keys at a time, Ruby would not be my first choice by far.
No, it doesn't. You ran 100k increments on one key and divided by time ONCE and printed the result.
Plus this was on your Macbook Pro, which is probably running a mail client, Twitter client, web browser, and hundreds of other things that could have affected the times.
Graphing the response times for each increment would be a good way to start looking at what the data really shows, instead of going by a single opaque, fairly useless number.
it could even be a dummy server just replying +OK and the result will not change.
You should try this against MySQL too, and you'll see similar numbers, for sure not an order of magnitude different than that even if SQL involves much more query processing and so forth.
For sure it seems like that MongoDB client library you are using has a smaller latency, perhaps it is a C library? As Redis-rb is instead written in Ruby itself.
Another huge problem of this benchmark is that it should try to increment a random key per every iteration, out of a dataset with a few millions of keys, as this is realistically what you need. You'll see how Redis with 50 clients will increment this counters 100k times per second or more without any performance degradation as the number of keys grows.