Since is way more expensive than a normal transaction, a node that is operating even at a fraction of its max TPS capacity might not be able to finish all requests for a block faster than a new block is created, even if running those calls in parallel.

Having a config that only traces failed TXs would be interesting because one of the more useful use-cases for having tracing enabled is knowing why a transaction failed, and since usually successful transactions are way more common than failed ones it would reduce the load on the node.