When switching from 2.2.1 driver to 2.4.1 certain queries started to take significantly longer.
I narrowed the problem down to the count method.

In configuration with a replica set, calls to .count take twice as long to process as they did pre-2.3 driver.
In tests with only one server, the problem was not presented. Due to this observation I believe the issue occurs somewhere in the server selection during count.
Other operations, find, where, do not exhibit the same behaviour.

I have attahed a file with two scripts, 2.2_test.sh and 2.3_test.sh.
Extract the archive on a ruby 2.2.4 environment with Bundler and you should be able to run the shell scripts to reproduce similar results as below:

This simple test, in an environment with 3 servers and an arbiter highlight the issue:
The only change in these two tests is modifying the Gemfile.lock mongo driver between 2.2.7 and 2.30:

(from the bundle console using the 2.2.7 driver)
> cli = Mongo::Client.new('mongodb://user:pass@server:11701/database')
=> #<Mongo::Client:0x70176662644920 cluster=server:11701, server2:10987, server3:11015, server4:11701>
> Benchmark.measure { 1.upto(100) { cli.command(

) } }
=> #<Benchmark::Tms:0x007fa69591d5a0
@cstime=0.0,
@cutime=0.0,
@label="",
@real=1.6692429999820888,
@stime=0.010000000000000009,
@total=0.05999999999999983,
@utime=0.04999999999999982>

(from bundle console using the 2.3.0 driver)
> cli = Mongo::Client.new('mongodb://user:pass@server:11701/database')
=> #<Mongo::Client:0x70327742040800 cluster=server:11701, server2:10987, server3:11015, server4:11701>
production prod_demosphere@db-mongo-61[4] (main)> Benchmark.measure { 1.upto(100) { cli.command(

) } }
=> #<Benchmark::Tms:0x007fecdcca6810
@cstime=0.0,
@cutime=0.0,
@label="",
@real=3.175779999990482,
@stime=0.010000000000000009,
@total=0.0700000000000005,
@utime=0.0600000000000005>