Determine if the cache or lazy changes in #4231 were most responsible for speedup

[petebachant]

Could inform future design decisions.

#4231 seems to be the culprit for last big jump here:

[szy21]

Yes, I would be interested to know e.g. if caching ᶜmixing_length will make the model faster.