Why are stress test results different?

Question

Nginx

And so, after some server setup, I decided to do a stress test.
Server front (so far 2) are in lxc containers, proxy nginx - on the host machine.
For the time being, I am not applying any settings for either nginx or manipulations with the kernel, if necessary, I will add.
In general, the test results are different and the behavior is strange. The test was done through siege.
siege my.site/some_url -r50 -c250
Here is one result

Transactions: 12451 hits
Availability: 99.61 %
Elapsed time: 36.84 secs
Data transferred: 80.52 MB
Response time: 0.06 secs
Transaction rate: 337.98 trans/sec
Throughput: 2.19 MB/sec
Concurrency: 21.43
Successful transactions: 12451
Failed transactions: 49
Longest transaction: 1.18
Shortest transaction: 0.04

And the next one with the same parameters

Transactions: 12152 hits
Availability: 97.22 %
Elapsed time: 200.90 secs
Data transferred: 29.16 MB
Response time: 1.15 secs
Transaction rate: 60.49 trans/sec
Throughput: 0.15 MB/sec
Concurrency: 69.67
Successful transactions: 12152
Failed transactions: 348
Longest transaction: 63.40
Shortest transaction: 0.16

If in the first case the indicators are good, then during the second the results are worse, and the behavior is strange, namely:
The test starts - the CPU of each container is loaded by 50%, but it falls over time, and the response time decreases. As if something is slowing it down, connection timeouts begin. Then again the speed increases, and so on in waves until the end of the test. During the test, I could not trace what this behavior is. Maybe someone has some ideas?
Thanks in advance for your advice ;)