In SERVER-46296 we had attempted to roll several builders over to run in fully dynamic mode. However, we discovered that this had a adverse effect on test execution time. It appears that a dynamically linked mongod is roughly 100 times slower than a statically linked mongod to start up.

It is probably impossible to get to complete parity with the static build, but we can almost certainly do better.

We need to figure out why the startup cost is so high, and bring it down to a level where we feel that the increased startup time is small enough that the benefits (much faster compiles) outweigh any residual test slowdown due to slower startup.