I ran some extensive testing in the Builder to find a good balance between strategy generation quality and time per accepted strategy. I then applied the resulting build config to the build task in my project. I noticed the time per accepted strategy using the same config in a project build task is significantly slower than when using the same config in the standalone builder. In my case it was 700ms in the Builder, and 2s in the Project Build task (Time per accepted strategy).
I know these times can vary per build run depending on many factors; however the above numbers are averages after 10 builds in both the builder and Project>Builder with each build stopping after 5000 strategies pass filters. So I think the differences might be significant.
Config attached for reference.
Using Random generation instead of genetic evolution will bring reproducible results, there is nothing that could cause this difference.