The data import used milliseconds instead of seconds, which overflowed the request_ts column. This breaks sort order, since there are a bunch of entries with the same, incorrect timestamp at the top of the list. (New entries are being written with the correct value, thus appear before the import.)
Additionally, the lack of filtering and pagination produces a very long list of profiles when following the "View in XHGui" link on a profiled page. The user has to find the relevant profile using the timestamp column, after waiting for the full list to load.