I'm all for the records breaking since it indicates to me there is increasing hashpower (miners) between each successive record breaker (3 2 birds with 1 stone). Obviously the data (with frequent updates) for hashpower peak optimization needed for superblock timing would be gathered from the pool that contributed the record breaker as the first tier filter then followed by the second, third, etc.

No prediction pattern can guarantee 100% accuracy but at least it's way better than having unacceptable hashpower (less than 80%) during the superblock event. Taking the drift assumption as reliable means of reaching the peak hashpower of any pool is far more less unpredictable. Inspecting the past few superblock records said it has no decisive & consistent direction.

Superblock   Date      Time      Drift      Drift from Average Time
160848   2014-11-17      19:50      0      -01:25
164880   2014-11-24      20:28      +00:38      -00:47
168912   2014-12-01      22:33      +02:05      +01:18
172944   2014-12-08      21:20      -01:13      +00:05
176976   2014-12-15      22:07      +00:47      +00:52
                     
   Average Time =      21:15