In v.1.0.14 we introduced automatic retry logic to fix some eventual consistency issues. The retries significantly affected the performance of destroy operations. In v1.0.16 we have added an option to disable the automatic retries. A limited number of resources had issues that were solved by the retry logic so we are adding this option as a short term fix. We are working on a better solution.
To avoid the longer time on destroy operations you can set the disable_auto_retries to true in the baremetal provider. See an example in docs/examples/compute/single_instance/provider.tf
Some people also reported they were seeing an error in some operating systems when a retry token was being used for multiple requests (See issue #170). v1.0.16 also includes a fix for this. We cannot test on every possible environment so please let us know if you are still seeing this issue when the automatic retries are enabled.