Inserting 10Ks of single tasks into the output queue (e.g., for a large LHS) can be slower than the time it takes to evaluate those tasks. So, for example, if there's a worker pool of 25K and 50K tasks are being inserted, the tasks may not be inserted fast enough to utilize the complete pool. This is especially the case in models with "short circuits" that may terminate quickly. A bulk insert to the output queue may be help with this, perhaps with some sort of chunking to match the worker pool size.
Tuning the worker pool size to account for task submissions time is also an important consideration here.
Inserting 10Ks of single tasks into the output queue (e.g., for a large LHS) can be slower than the time it takes to evaluate those tasks. So, for example, if there's a worker pool of 25K and 50K tasks are being inserted, the tasks may not be inserted fast enough to utilize the complete pool. This is especially the case in models with "short circuits" that may terminate quickly. A bulk insert to the output queue may be help with this, perhaps with some sort of chunking to match the worker pool size.
Tuning the worker pool size to account for task submissions time is also an important consideration here.