@AmandaAskell makes sense, maybe the ratios being off might cause a bottleneck issue that creates overhead that worsens the bottleneck similar to this:

Posters: visakanv. visakanv

2 likes recorded by the source corpus.