No, their tests don't imply that they are getting better write times with redis across the board, they are measuring the 99%-tile latency. What they have found

No, their tests don't imply that they are getting better write times with redis across the board, they are measuring the 99%-tile latency. What they have found, in a worst case scenario Redis is worse than Postgres. So for example, their Redis response times might look like ,{1ms, 1ms, 1ms, 9ms}, while their Postgres response times might look like {3ms, 4ms, 3ms, 4ms}. Looking at it this way you can expect redis to perform better, however that outlier is worrying. For stripe they value having consistent and predictable performance over varying performance.