The "time" result should be taken from across multiple requests and not a single request

[motin]

Currently, the "time" results is taken from a single request

I suggest we use the result that ab already reports "Time per request: 78.445 ms" which is the mean based on all requests. It would also be interesting to use ab's "Time per request: 7.845 [ms](mean, across all concurrent requests)" for comparison reasons.

[kenjis]

Yeah, I also want to fix the problem. Once is too less, and the value is not stable.
But ab's time is not the same.

[motin]

What about taking ab's time for each framework and then subtracting ab's time for "no-framework" requests? That would give the actual real-world additional bootstrap cost for the framework. Related discussion: #34