Some things to consider:

• Does the page server ever starve the log server
• Consider creating an intense workload where the page server is not reading from wt cache AND the oplog is very large so the log server needs an aggressive number of tokens.
• Is it better to bound boundaries by available tokens as opposed to waiting for the available tokens to let the 1MB batch go through?
• The batch is generated before we try to acquire the tokens, maybe something we can check for when generating batches?