Performance regression in replace.Convert.time_replace

[TomAugspurger]

#34999 caused a 2x slowdown in replace.Convert.time_replace (https://pandas.pydata.org/speed/pandas/index.html#replace.Convert.time_replace?p-constructor=%27Series%27&p-replace_data=%27Timestamp%27&commits=dbfbef7eaea7f23a42d9a65344b5745fc055f5b4)

[jbrockmendel]

How do you get 100x? do you mean 100ms?

[TomAugspurger]

Sorry, meant 100% (2x). Edited.

[jbrockmendel]

Assuming this was caused by#34999, this is something we just have to live with. The previous behavior was incorrect.

[TomAugspurger]

Perhaps a fluke in the timing? The latest runs are back to normal.